diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000000000000000000000000000000000000..a6344aac8c09253b3b630fb776ae94478aa0275b --- /dev/null +++ b/.gitattributes @@ -0,0 +1,35 @@ +*.7z filter=lfs diff=lfs merge=lfs -text +*.arrow filter=lfs diff=lfs merge=lfs -text +*.bin filter=lfs diff=lfs merge=lfs -text +*.bz2 filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text +*.ftz filter=lfs diff=lfs merge=lfs -text +*.gz filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.joblib filter=lfs diff=lfs merge=lfs -text +*.lfs.* filter=lfs diff=lfs merge=lfs -text +*.mlmodel filter=lfs diff=lfs merge=lfs -text +*.model filter=lfs diff=lfs merge=lfs -text +*.msgpack filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.onnx filter=lfs diff=lfs merge=lfs -text +*.ot filter=lfs diff=lfs merge=lfs -text +*.parquet filter=lfs diff=lfs merge=lfs -text +*.pb filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text +*.pt filter=lfs diff=lfs merge=lfs -text +*.pth filter=lfs diff=lfs merge=lfs -text +*.rar filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +saved_model/**/* filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.tar filter=lfs diff=lfs merge=lfs -text +*.tflite filter=lfs diff=lfs merge=lfs -text +*.tgz filter=lfs diff=lfs merge=lfs -text +*.wasm filter=lfs diff=lfs merge=lfs -text +*.xz filter=lfs diff=lfs merge=lfs -text +*.zip filter=lfs diff=lfs merge=lfs -text +*.zst filter=lfs diff=lfs merge=lfs -text +*tfevents* filter=lfs diff=lfs merge=lfs -text diff --git a/.gitignore b/.gitignore new file mode 100644 index 0000000000000000000000000000000000000000..fb4002e5c74fad144ac64cfb3c4010b2152a8152 --- /dev/null +++ b/.gitignore @@ -0,0 +1,6 @@ +# main/evaluation +.vscode/ + +*.pyc +gradio_temp +*.pth \ No newline at end of file diff --git a/LICENSE b/LICENSE new file mode 100644 index 0000000000000000000000000000000000000000..261eeb9e9f8b2b4b0d119366dda99c6fd7d35c64 --- /dev/null +++ b/LICENSE @@ -0,0 +1,201 @@ + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright [yyyy] [name of copyright owner] + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. diff --git a/README.md b/README.md new file mode 100644 index 0000000000000000000000000000000000000000..0eeeb5195d2acd815cb8aaebdffe39689ed01f83 --- /dev/null +++ b/README.md @@ -0,0 +1,83 @@ + +
+ +

MotionCtrl: A Unified and Flexible + Motion Controller + for Video Generation

+ + + +[![ Paper](https://img.shields.io/badge/Paper-gray +)](https://wzhouxiff.github.io/projects/MotionCtrl/assets/paper/MotionCtrl.pdf)   [![ arXiv](https://img.shields.io/badge/arXiv-red +)](https://arxiv.org/pdf/2312.03641.pdf)   [![Porject Page](https://img.shields.io/badge/Project%20Page-green +) +](https://wzhouxiff.github.io/projects/MotionCtrl/)   [![Demo](https://img.shields.io/badge/Gradio%20Demo-orange +)]() + +[Zhouxia Wang](https://vvictoryuki.github.io/website/)1,2, [Ziyang Yuan](https://github.com/jiangyzy)1,4, [Xintao Wang](https://xinntao.github.io/)1,3, [Tianshui Chen](http://tianshuichen.com/)6, [Menghan Xia](https://menghanxia.github.io/)3, [Ping Luo](http://luoping.me/)2,5, [Ying Shan](https://scholar.google.com/citations?hl=zh-CN&user=4oXBp9UAAAAJ)1,3 + +1 ARC Lab, Tencent PCG, 2 The University of Hong Kong, 3 Tencent AI Lab, 4 Tsinghua University, 5 Shanghai AI Laboratory, 6 Guangdong University of Technology + + +
+ + + +Our proposed MotionCtrl is capable of independently controlling the complex camera motion and object motion of the generated videos, with only a unified model. +There are some results attained with MotionCtrl and more results are showcased in our [Project Page](https://wzhouxiff.github.io/projects/MotionCtrl/). + + + +
+ + + + +
+ + +## Updating +- [x] Release MotionCtrl depolyed on *LVDM/VideoCrafter* +- [ ] Gradio Demo Available + + + +## Inference + +1. Download the weights of MotionCtrl from [weipan](https://drive.weixin.qq.com/s?k=AJEAIQdfAAogLtIAPh) to `./checkpoints`. +2. Go into `configs/inference/run.sh` and set `condtype` as 'camera_motion', 'object_motion', or 'both'. +- `condtype=camera_motion` means only control the **camera motion** in the generated video. +- `condtype=object_motion` means only control the **object motion** in the generated video. +- `condtype=both` means control the camera motion and object motion in the generated video **simultaneously**. +3. sh configs/inference/run.sh + +## Citation +If you make use of our work, please cite our paper. +```bibtex +@inproceedings{wang2023motionctrl, + title={MotionCtrl: A Unified and Flexible Motion Controller for Video Generation}, + author={Wang, Zhouxia and Yuan, Ziyang and Wang, Xintao and Chen, Tianshui and Xia, Menghan and Luo, Ping and Shan, Yin}, + booktitle={arXiv preprint arXiv:2312.03641}, + year={2023} +} +``` + diff --git a/app.py b/app.py new file mode 100644 index 0000000000000000000000000000000000000000..90185aba6de58391d227daaeea9fe8e9ccd568f0 --- /dev/null +++ b/app.py @@ -0,0 +1,1079 @@ +import argparse +import os +import tempfile +from functools import partial + +import cv2 +import gradio as gr +import imageio +import numpy as np +import torch +import torchvision +from omegaconf import OmegaConf +from PIL import Image +from pytorch_lightning import seed_everything + +from gradio_utils.camera_utils import CAMERA_MOTION_MODE, process_camera +from gradio_utils.traj_utils import (OBJECT_MOTION_MODE, get_provided_traj, + process_points, process_traj) +from gradio_utils.utils import vis_camera +from lvdm.models.samplers.ddim import DDIMSampler +from main.evaluation.motionctrl_inference import (DEFAULT_NEGATIVE_PROMPT, + load_model_checkpoint, + post_prompt) +from utils.utils import instantiate_from_config + +os.environ['KMP_DUPLICATE_LIB_OK']='True' + + +#### Description #### +title = r"""

MotionCtrl: A Unified and Flexible Motion Controller for Video Generation

""" + +description = r""" +Official Gradio demo for MotionCtrl: A Unified and Flexible Motion Controller for Video Generation.
+🔥 MotionCtrl is capable of independently and flexibly controling the camera motion and object motion of a generated video, with only a unified model.
+🤗 Try to control the motion of the generated videos yourself!
+❗❗❗ Please note that current version of **MotionCtrl** is deployed on **LVDM/VideoCrafter**. The versions that depolyed on **AnimateDiff** and **SVD** will be released soon.
+""" +article = r""" +If MotionCtrl is helpful, please help to ⭐ the Github Repo. Thanks! +[![GitHub Stars](https://img.shields.io/github/stars/TencentARC%2FMotionCtrl +)](https://github.com/TencentARC/MotionCtrl) + +--- + +📝 **Citation** +
+If our work is useful for your research, please consider citing: +```bibtex +@inproceedings{wang2023motionctrl, + title={MotionCtrl: A Unified and Flexible Motion Controller for Video Generation}, + author={Wang, Zhouxia and Yuan, Ziyang and Wang, Xintao and Chen, Tianshui and Xia, Menghan and Luo, Ping and Shan, Yin}, + booktitle={arXiv preprint arXiv:2312.03641}, + year={2023} +} +``` + +📧 **Contact** +
+If you have any questions, please feel free to reach me out at wzhoux@connect.hku.hk. + +""" +css = """ +.gradio-container {width: 85% !important} +.gr-monochrome-group {border-radius: 5px !important; border: revert-layer !important; border-width: 2px !important; color: black !important;} +span.svelte-s1r2yt {font-size: 17px !important; font-weight: bold !important; color: #d30f2f !important;} +button {border-radius: 8px !important;} +.add_button {background-color: #4CAF50 !important;} +.remove_button {background-color: #f44336 !important;} +.clear_button {background-color: gray !important;} +.mask_button_group {gap: 10px !important;} +.video {height: 300px !important;} +.image {height: 300px !important;} +.video .wrap.svelte-lcpz3o {display: flex !important; align-items: center !important; justify-content: center !important;} +.video .wrap.svelte-lcpz3o > :first-child {height: 100% !important;} +.margin_center {width: 50% !important; margin: auto !important;} +.jc_center {justify-content: center !important;} +""" + + +T_base = [ + [1.,0.,0.], ## W2C x 的正方向： 相机朝左 left + [-1.,0.,0.], ## W2C x 的负方向： 相机朝右 right + [0., 1., 0.], ## W2C y 的正方向： 相机朝上 up + [0.,-1.,0.], ## W2C y 的负方向： 相机朝下 down + [0.,0.,1.], ## W2C z 的正方向： 相机往前 zoom out + [0.,0.,-1.], ## W2C z 的负方向： 相机往前 zoom in + ] +radius = 1 +n = 16 +# step = +look_at = np.array([0, 0, 0.8]).reshape(3,1) +# look_at = np.array([0, 0, 0.2]).reshape(3,1) + +T_list = [] +base_R = np.array([[1., 0., 0.], + [0., 1., 0.], + [0., 0., 1.]]) +res = [] +res_forsave = [] +T_range = 1.8 + + + +for i in range(0, 16): + # theta = (1)*np.pi*i/n + + R = base_R[:,:3] + T = np.array([0.,0.,1.]).reshape(3,1) * (i/n)*2 + RT = np.concatenate([R,T], axis=1) + res.append(RT) + +fig = vis_camera(res) + +# MODE = ["camera motion control", "object motion control", "camera + object motion control"] +MODE = ["control camera poses", "control object trajectory", "control both camera and object motion"] +BASE_MODEL = ['LVDM/VideoCrafter', 'AnimateDiff', 'SVD'] + + +traj_list = [] +camera_dict = { + "motion":[], + "mode": "Customized Mode 1: First A then B", # "First A then B", "Both A and B", "Custom" + "speed": 1.0, + "complex": None + } + +def fn_vis_camera(info_mode): + global camera_dict + RT = process_camera(camera_dict) # [t, 3, 4] + if camera_dict['complex'] is not None: + # rescale T to [-2,2] + for i in range(3): + min_T = np.min(RT[:,i,-1]) + max_T = np.max(RT[:,i,-1]) + if min_T < -2 or max_T > 2: + RT[:,i,-1] = RT[:,i,-1] - min_T + RT[:,i,-1] = RT[:,i,-1] / (np.max(RT[:,:,-1]) + 1e-6) + RT[:,i,-1] = RT[:,i,-1] * 4 + RT[:,i,-1] = RT[:,i,-1] - 2 + + fig = vis_camera(RT) + + if info_mode == MODE[0]: + vis_step3_prompt_generate = True + vis_prompt = True + vis_num_samples = True + vis_seed = True + vis_start = True + vis_gen_video = True + + vis_object_mode = False + vis_object_info = False + + else: + vis_step3_prompt_generate = False + vis_prompt = False + vis_num_samples = False + vis_seed = False + vis_start = False + vis_gen_video = False + + vis_object_mode = True + vis_object_info = True + + return fig, \ + gr.update(visible=vis_object_mode), \ + gr.update(visible=vis_object_info), \ + gr.update(visible=vis_step3_prompt_generate), \ + gr.update(visible=vis_prompt), \ + gr.update(visible=vis_num_samples), \ + gr.update(visible=vis_seed), \ + gr.update(visible=vis_start), \ + gr.update(visible=vis_gen_video, value=None) + +def fn_vis_traj(): + global traj_list + xy_range = 1024 + points = process_points(traj_list) + imgs = [] + for idx in range(16): + bg_img = np.ones((1024, 1024, 3), dtype=np.uint8) * 255 + for i in range(15): + p = points[i] + p1 = points[i+1] + cv2.line(bg_img, p, p1, (255, 0, 0), 2) + + if i == idx: + cv2.circle(bg_img, p, 2, (0, 255, 0), 20) + + if idx==(15): + cv2.circle(bg_img, points[-1], 2, (0, 255, 0), 20) + + imgs.append(bg_img.astype(np.uint8)) + + # size = (512, 512) + fps = 10 + path = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False).name + writer = imageio.get_writer(path, format='mp4', mode='I', fps=fps) + for img in imgs: + writer.append_data(img) + + writer.close() + + vis_step3_prompt_generate = True + vis_prompt = True + vis_num_samples = True + vis_seed = True + vis_start = True + vis_gen_video = True + return path, gr.update(visible=vis_step3_prompt_generate), \ + gr.update(visible=vis_prompt), \ + gr.update(visible=vis_num_samples), \ + gr.update(visible=vis_seed), \ + gr.update(visible=vis_start), \ + gr.update(visible=vis_gen_video, value=None) + +def display_camera_info(camera_dict, camera_mode=None): + if camera_dict['complex'] is not None: + res = f"complex : {camera_dict['complex']}. " + else: + res = "" + res += f"motion : {[_ for _ in camera_dict['motion']]}. " + res += f"speed : {camera_dict['speed']}. " + if camera_mode == 'Custom Camera Poses': + res += f"mode : {camera_dict['mode']}. " + return res + +def add_traj_point(evt: gr.SelectData, ): + global traj_list + traj_list.append(evt.index) + traj_str = [f"{traj}" for traj in traj_list] + return ", ".join(traj_str) + +def add_provided_traj(traj_name): + global traj_list + traj_list = get_provided_traj(traj_name) + traj_str = [f"{traj}" for traj in traj_list] + return ", ".join(traj_str) + +def add_camera_motion(camera_motion, camera_mode): + global camera_dict + if camera_dict['complex'] is not None: + camera_dict['complex'] = None + if camera_mode == 'Custom Camera Poses' and len(camera_dict['motion']) <2: + camera_dict['motion'].append(camera_motion) + else: + camera_dict['motion']=[camera_motion] + + return display_camera_info(camera_dict, camera_mode) + +def add_complex_camera_motion(camera_motion): + global camera_dict + camera_dict['complex']=camera_motion + return display_camera_info(camera_dict) + +def change_camera_mode(combine_type, camera_mode): + global camera_dict + camera_dict['mode'] = combine_type + + return display_camera_info(camera_dict, camera_mode) + +def change_camera_speed(camera_speed): + global camera_dict + camera_dict['speed'] = camera_speed + return display_camera_info(camera_dict) + +def reset_camera(): + global camera_dict + camera_dict = { + "motion":[], + "mode": "Customized Mode 1: First A then B", + "speed": 1.0, + "complex": None + } + return display_camera_info(camera_dict) + + +def fn_traj_droplast(): + global traj_list + + if traj_list: + traj_list.pop() + + if traj_list: + traj_str = [f"{traj}" for traj in traj_list] + return ", ".join(traj_str) + else: + return "Click to specify trajectory" + +def fn_traj_reset(): + global traj_list + traj_list = [] + return "Click to specify trajectory" + +########################################### +model_path='./checkpoints/motionctrl.pth' +config_path='./configs/inference/config_both.yaml' + +config = OmegaConf.load(config_path) +model_config = config.pop("model", OmegaConf.create()) +model = instantiate_from_config(model_config) +model = model.cuda() + +model = load_model_checkpoint(model, model_path) +model.eval() + + +def model_run(prompts, infer_mode, seed, n_samples): + global traj_list + global camera_dict + + RT = process_camera(camera_dict).reshape(-1,12) + traj_flow = process_traj(traj_list).transpose(3,0,1,2) + print(prompts) + print(RT.shape) + print(traj_flow.shape) + + noise_shape = [1, 4, 16, 32, 32] + unconditional_guidance_scale = 7.5 + unconditional_guidance_scale_temporal = None + # n_samples = 1 + ddim_steps= 50 + ddim_eta=1.0 + cond_T=800 + + if n_samples < 1: + n_samples = 1 + if n_samples > 4: + n_samples = 4 + + seed_everything(seed) + + if infer_mode == MODE[0]: + camera_poses = RT + camera_poses = torch.tensor(camera_poses).float().cuda() + camera_poses = camera_poses.unsqueeze(0) + trajs = None + elif infer_mode == MODE[1]: + trajs = traj_flow + trajs = torch.tensor(trajs).float().cuda() + trajs = trajs.unsqueeze(0) + camera_poses = None + else: + camera_poses = RT + trajs = traj_flow + camera_poses = torch.tensor(camera_poses).float().cuda() + trajs = torch.tensor(trajs).float().cuda() + camera_poses = camera_poses.unsqueeze(0) + trajs = trajs.unsqueeze(0) + + ddim_sampler = DDIMSampler(model) + batch_size = noise_shape[0] + ## get condition embeddings (support single prompt only) + if isinstance(prompts, str): + prompts = [prompts] + + for i in range(len(prompts)): + prompts[i] = f'{prompts[i]}, {post_prompt}' + + cond = model.get_learned_conditioning(prompts) + if camera_poses is not None: + RT = camera_poses[..., None] + else: + RT = None + + if trajs is not None: + traj_features = model.get_traj_features(trajs) + else: + traj_features = None + + if unconditional_guidance_scale != 1.0: + # prompts = batch_size * [""] + prompts = batch_size * [DEFAULT_NEGATIVE_PROMPT] + uc = model.get_learned_conditioning(prompts) + if traj_features is not None: + un_motion = model.get_traj_features(torch.zeros_like(trajs)) + else: + un_motion = None + uc = {"features_adapter": un_motion, "uc": uc} + else: + uc = None + + batch_variants = [] + for _ in range(n_samples): + if ddim_sampler is not None: + samples, _ = ddim_sampler.sample(S=ddim_steps, + conditioning=cond, + batch_size=noise_shape[0], + shape=noise_shape[1:], + verbose=False, + unconditional_guidance_scale=unconditional_guidance_scale, + unconditional_conditioning=uc, + eta=ddim_eta, + temporal_length=noise_shape[2], + conditional_guidance_scale_temporal=unconditional_guidance_scale_temporal, + features_adapter=traj_features, + pose_emb=RT, + cond_T=cond_T + ) + ## reconstruct from latent to pixel space + batch_images = model.decode_first_stage(samples) + batch_variants.append(batch_images) + ## variants, batch, c, t, h, w + batch_variants = torch.stack(batch_variants, dim=1) + batch_variants = batch_variants[0] + + # file_path = save_results(batch_variants, "MotionCtrl", "gradio_temp", fps=10) + file_path = save_results(batch_variants, fps=10) + print(file_path) + + return gr.update(value=file_path, width=256*n_samples, height=256) + + # return file_path + +def save_results(video, fps=10): + + # b,c,t,h,w + video = video.detach().cpu() + video = torch.clamp(video.float(), -1., 1.) + n = video.shape[0] + video = video.permute(2, 0, 1, 3, 4) # t,n,c,h,w + frame_grids = [torchvision.utils.make_grid(framesheet, nrow=int(n)) for framesheet in video] #[3, 1*h, n*w] + grid = torch.stack(frame_grids, dim=0) # stack in temporal dim [t, 3, n*h, w] + grid = (grid + 1.0) / 2.0 + grid = (grid * 255).to(torch.uint8).permute(0, 2, 3, 1) # [t, h, w*n, 3] + + path = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False).name + + writer = imageio.get_writer(path, format='mp4', mode='I', fps=fps) + for i in range(grid.shape[0]): + img = grid[i].numpy() + writer.append_data(img) + + writer.close() + + return path + +def visualized_step2(infer_mode): + + # reset + reset_camera() + fn_traj_reset() + + # camera motion control + vis_basic_camera_motion = False + vis_basic_camera_motion_des = False + vis_custom_camera_motion = False + vis_custom_run_status = False + vis_complex_camera_motion = False + vis_complex_camera_motion_des = False + vis_U = False + vis_D = False + vis_L = False + vis_R = False + vis_I = False + vis_O = False + vis_ACW = False + vis_CW = False + vis_combine1 = False + vis_combine2 = False + vis_speed = False + + vis_Pose_1, vis_Pose_2, vis_Pose_3, vis_Pose_4 = False, False, False, False + vis_Pose_5, vis_Pose_6, vis_Pose_7, vis_Pose_8 = False, False, False, False + + vis_camera_args = False + vis_camera_reset = False + vis_camera_vis = False + vis_vis_camera = False + + # object motion control + vis_provided_traj = False + vis_provided_traj_des = False + vis_draw_yourself = False + vis_draw_run_status = False + + vis_traj_1, vis_traj_2, vis_traj_3, vis_traj_4 = False, False, False, False + vis_traj_5, vis_traj_6, vis_traj_7, vis_traj_8 = False, False, False, False + + traj_args = False + traj_droplast, traj_reset = False, False + traj_vis = False + traj_input, vis_traj = False, False + + + # generate video + vis_step3_prompt_generate = False + vis_prompt = False + vis_num_samples = False + vis_seed = False + vis_start = False + vis_gen_video = False + + if infer_mode == MODE[0]: + vis_step2_camera_motion = True + vis_step2_camera_motion_des = True + vis_camera_mode = True + vis_camera_info = True + + vis_step2_object_motion = False + vis_step2_object_motion_des = False + vis_traj_mode = False + vis_traj_info = False + + step2_camera_object_motion = False + step2_camera_object_motion_des = False + + elif infer_mode == MODE[1]: + vis_step2_camera_motion = False + vis_step2_camera_motion_des = False + vis_camera_mode = False + vis_camera_info = False + + vis_step2_object_motion = True + vis_step2_object_motion_des = True + vis_traj_mode = True + vis_traj_info = True + + step2_camera_object_motion = False + step2_camera_object_motion_des = False + else: #infer_mode == MODE[2]: + vis_step2_camera_motion = False + vis_step2_camera_motion_des = False + vis_camera_mode = False + vis_camera_info = False + + vis_step2_object_motion = False + vis_step2_object_motion_des = False + vis_traj_mode = False + vis_traj_info = False + + step2_camera_object_motion = True + step2_camera_object_motion_des = True + + vis_basic_camera_motion = True + vis_basic_camera_motion_des = True + vis_U = True + vis_D = True + vis_L = True + vis_R = True + vis_I = True + vis_O = True + vis_ACW = True + vis_CW = True + vis_speed = True + + vis_camera_args = True + vis_camera_reset = True + vis_camera_vis = True + vis_vis_camera = True + + + return gr.update(visible=vis_step2_camera_motion), \ + gr.update(visible=vis_step2_camera_motion_des), \ + gr.update(visible=vis_camera_mode), \ + gr.update(visible=vis_camera_info), \ + gr.update(visible=vis_basic_camera_motion), \ + gr.update(visible=vis_basic_camera_motion_des), \ + gr.update(visible=vis_custom_camera_motion), \ + gr.update(visible=vis_custom_run_status), \ + gr.update(visible=vis_complex_camera_motion), \ + gr.update(visible=vis_complex_camera_motion_des), \ + gr.update(visible=vis_U), gr.update(visible=vis_D), gr.update(visible=vis_L), gr.update(visible=vis_R), \ + gr.update(visible=vis_I), gr.update(visible=vis_O), gr.update(visible=vis_ACW), gr.update(visible=vis_CW), \ + gr.update(visible=vis_combine1), gr.update(visible=vis_combine2), \ + gr.update(visible=vis_speed), \ + gr.update(visible=vis_Pose_1), gr.update(visible=vis_Pose_2), gr.update(visible=vis_Pose_3), gr.update(visible=vis_Pose_4), \ + gr.update(visible=vis_Pose_5), gr.update(visible=vis_Pose_6), gr.update(visible=vis_Pose_7), gr.update(visible=vis_Pose_8), \ + gr.update(visible=vis_camera_args, value=None), \ + gr.update(visible=vis_camera_reset), gr.update(visible=vis_camera_vis), \ + gr.update(visible=vis_vis_camera, value=None), \ + gr.update(visible=vis_step2_object_motion), \ + gr.update(visible=vis_step2_object_motion_des), \ + gr.update(visible=vis_traj_mode), \ + gr.update(visible=vis_traj_info), \ + gr.update(visible=vis_provided_traj), \ + gr.update(visible=vis_provided_traj_des), \ + gr.update(visible=vis_draw_yourself), \ + gr.update(visible=vis_draw_run_status), \ + gr.update(visible=vis_traj_1), gr.update(visible=vis_traj_2), gr.update(visible=vis_traj_3), gr.update(visible=vis_traj_4), \ + gr.update(visible=vis_traj_5), gr.update(visible=vis_traj_6), gr.update(visible=vis_traj_7), gr.update(visible=vis_traj_8), \ + gr.update(visible=traj_args), \ + gr.update(visible=traj_droplast), gr.update(visible=traj_reset), \ + gr.update(visible=traj_vis), \ + gr.update(visible=traj_input), gr.update(visible=vis_traj, value=None), \ + gr.update(visible=step2_camera_object_motion), \ + gr.update(visible=step2_camera_object_motion_des), \ + gr.update(visible=vis_step3_prompt_generate), \ + gr.update(visible=vis_prompt), \ + gr.update(visible=vis_num_samples), \ + gr.update(visible=vis_seed), \ + gr.update(visible=vis_start), \ + gr.update(visible=vis_gen_video) + +def visualized_camera_poses(step2_camera_motion): + reset_camera() + + # generate video + vis_step3_prompt_generate = False + vis_prompt = False + vis_num_samples = False + vis_seed = False + vis_start = False + vis_gen_video = False + + if step2_camera_motion == CAMERA_MOTION_MODE[0]: + vis_basic_camera_motion = True + vis_basic_camera_motion_des = True + vis_custom_camera_motion = False + vis_custom_run_status = False + vis_complex_camera_motion = False + vis_complex_camera_motion_des = False + vis_U = True + vis_D = True + vis_L = True + vis_R = True + vis_I = True + vis_O = True + vis_ACW = True + vis_CW = True + vis_combine1 = False + vis_combine2 = False + vis_speed = True + + vis_Pose_1, vis_Pose_2, vis_Pose_3, vis_Pose_4 = False, False, False, False + vis_Pose_5, vis_Pose_6, vis_Pose_7, vis_Pose_8 = False, False, False, False + + elif step2_camera_motion == CAMERA_MOTION_MODE[1]: + vis_basic_camera_motion = False + vis_basic_camera_motion_des = False + vis_custom_camera_motion = False + vis_custom_run_status = False + vis_complex_camera_motion = True + vis_complex_camera_motion_des = True + vis_U = False + vis_D = False + vis_L = False + vis_R = False + vis_I = False + vis_O = False + vis_ACW = False + vis_CW = False + vis_combine1 = False + vis_combine2 = False + vis_speed = False + + vis_Pose_1, vis_Pose_2, vis_Pose_3, vis_Pose_4 = True, True, True, True + vis_Pose_5, vis_Pose_6, vis_Pose_7, vis_Pose_8 = True, True, True, True + + else: # step2_camera_motion = CAMERA_MOTION_MODE[2]: + vis_basic_camera_motion = False + vis_basic_camera_motion_des = False + vis_custom_camera_motion = True + vis_custom_run_status = True + vis_complex_camera_motion = False + vis_complex_camera_motion_des = False + vis_U = True + vis_D = True + vis_L = True + vis_R = True + vis_I = True + vis_O = True + vis_ACW = True + vis_CW = True + vis_combine1 = True + vis_combine2 = True + vis_speed = True + + vis_Pose_1, vis_Pose_2, vis_Pose_3, vis_Pose_4 = False, False, False, False + vis_Pose_5, vis_Pose_6, vis_Pose_7, vis_Pose_8 = False, False, False, False + + vis_camera_args = True + vis_camera_reset = True + vis_camera_vis = True + vis_vis_camera = True + + return gr.update(visible=vis_basic_camera_motion), \ + gr.update(visible=vis_basic_camera_motion_des), \ + gr.update(visible=vis_custom_camera_motion), \ + gr.update(visible=vis_custom_run_status), \ + gr.update(visible=vis_complex_camera_motion), \ + gr.update(visible=vis_complex_camera_motion_des), \ + gr.update(visible=vis_U), gr.update(visible=vis_D), gr.update(visible=vis_L), gr.update(visible=vis_R), \ + gr.update(visible=vis_I), gr.update(visible=vis_O), gr.update(visible=vis_ACW), gr.update(visible=vis_CW), \ + gr.update(visible=vis_combine1), gr.update(visible=vis_combine2), \ + gr.update(visible=vis_speed), \ + gr.update(visible=vis_Pose_1), gr.update(visible=vis_Pose_2), gr.update(visible=vis_Pose_3), gr.update(visible=vis_Pose_4), \ + gr.update(visible=vis_Pose_5), gr.update(visible=vis_Pose_6), gr.update(visible=vis_Pose_7), gr.update(visible=vis_Pose_8), \ + gr.update(visible=vis_camera_args, value=None), \ + gr.update(visible=vis_camera_reset), gr.update(visible=vis_camera_vis), \ + gr.update(visible=vis_vis_camera, value=None), \ + gr.update(visible=vis_step3_prompt_generate), \ + gr.update(visible=vis_prompt), \ + gr.update(visible=vis_num_samples), \ + gr.update(visible=vis_seed), \ + gr.update(visible=vis_start), \ + gr.update(visible=vis_gen_video) + +def visualized_traj_poses(step2_object_motion): + + fn_traj_reset() + + # generate video + vis_step3_prompt_generate = False + vis_prompt = False + vis_num_samples = False + vis_seed = False + vis_start = False + vis_gen_video = False + + if step2_object_motion == "Provided Trajectory": + vis_provided_traj = True + vis_provided_traj_des = True + vis_draw_yourself = False + vis_draw_run_status = False + + vis_traj_1, vis_traj_2, vis_traj_3, vis_traj_4 = True, True, True, True + vis_traj_5, vis_traj_6, vis_traj_7, vis_traj_8 = True, True, True, True + + traj_args = True + traj_droplast, traj_reset = False, True + traj_vis = True + traj_input, vis_traj = False, True + + + elif step2_object_motion == "Custom Trajectory": + vis_provided_traj = False + vis_provided_traj_des = False + vis_draw_yourself = True + vis_draw_run_status = True + + vis_traj_1, vis_traj_2, vis_traj_3, vis_traj_4 = False, False, False, False + vis_traj_5, vis_traj_6, vis_traj_7, vis_traj_8 = False, False, False, False + + traj_args = True + traj_droplast, traj_reset = True, True + traj_vis = True + traj_input, vis_traj = True, True + + return gr.update(visible=vis_provided_traj), \ + gr.update(visible=vis_provided_traj_des), \ + gr.update(visible=vis_draw_yourself), \ + gr.update(visible=vis_draw_run_status), \ + gr.update(visible=vis_traj_1), gr.update(visible=vis_traj_2), gr.update(visible=vis_traj_3), gr.update(visible=vis_traj_4), \ + gr.update(visible=vis_traj_5), gr.update(visible=vis_traj_6), gr.update(visible=vis_traj_7), gr.update(visible=vis_traj_8), \ + gr.update(visible=traj_args), \ + gr.update(visible=traj_droplast), gr.update(visible=traj_reset), \ + gr.update(visible=traj_vis), \ + gr.update(visible=traj_input), gr.update(visible=vis_traj, value=None), \ + gr.update(visible=vis_step3_prompt_generate), \ + gr.update(visible=vis_prompt), \ + gr.update(visible=vis_num_samples), \ + gr.update(visible=vis_seed), \ + gr.update(visible=vis_start), \ + gr.update(visible=vis_gen_video) + +def main(args): + demo = gr.Blocks() + with demo: + + gr.Markdown(title) + gr.Markdown(description) + + # state = gr.State({ + # "mode": "camera_only", + # "camera_input": [], + # "traj_input": [], + # }) + + with gr.Column(): + ''' + # step 0: select based model. + gr.Markdown("## Step0: Selecting the model", show_label=False) + gr.Markdown( f'- {BASE_MODEL[0]}: **MotionCtrl** deployed on {BASE_MODEL[0]}', show_label=False) + gr.Markdown( f'- {BASE_MODEL[1]}: **MotionCtrl** deployed on {BASE_MODEL[1]}', show_label=False) + gr.Markdown( f'- {BASE_MODEL[2]}: **MotionCtrl** deployed on {BASE_MODEL[2]}', show_label=False) + gr.Markdown( f'- **Only the model that deployed on {BASE_MODEL[0]} is avalible now. MotionCtrl models deployed on {BASE_MODEL[1]} and {BASE_MODEL[2]} are coming soon.**', show_label=False) + gr.Radio(choices=BASE_MODEL, value=BASE_MODEL[0], label="Based Model", interactive=False) + ''' + + # step 1: select motion control mode + gr.Markdown("## Step 1/3: Selecting the motion control mode", show_label=False) + gr.Markdown( f'- {MODE[0]}: Control the camera motion only', show_label=False) + gr.Markdown( f'- {MODE[1]}: Control the object motion only', show_label=False) + gr.Markdown( f'- {MODE[2]}: Control both the camera and object motion', show_label=False) + gr.Markdown( f'- Click `Proceed` to go into next step', show_label=False) + infer_mode = gr.Radio(choices=MODE, value=MODE[0], label="Motion Control Mode", interactive=True) + mode_info = gr.Button(value="Proceed") + + # step2 - camera + object motion control + step2_camera_object_motion = gr.Markdown("---\n## Step 2/3: Select the camera poses and trajectory", show_label=False, visible=False) + step2_camera_object_motion_des = gr.Markdown(f"\n 1. Select a basic camera pose. \ + \n 2. Select a provided trajectory or draw the trajectory yourself.", + show_label=False, visible=False) + + # step2 - camera motion control + step2_camera_motion = gr.Markdown("---\n## Step 2/3: Select the camera poses", show_label=False, visible=False) + step2_camera_motion_des = gr.Markdown(f"\n - {CAMERA_MOTION_MODE[0]}: Including 8 basic camera poses, such as pan up, pan down, zoom in, and zoom out. \ + \n - {CAMERA_MOTION_MODE[1]}: Complex camera poses extracted from the real videos. \ + \n - {CAMERA_MOTION_MODE[2]}: You can customize complex camera poses yourself by combining or fusing two of the eight basic camera poses. \ + \n - Click `Proceed` to go into next step", + show_label=False, visible=False) + camera_mode = gr.Radio(choices=CAMERA_MOTION_MODE, value=CAMERA_MOTION_MODE[0], label="Camera Motion Control Mode", interactive=True, visible=False) + camera_info = gr.Button(value="Proceed", visible=False) + + with gr.Row(): + with gr.Column(): + # step2.1 - camera motion control - basic + basic_camera_motion = gr.Markdown("---\n### Basic Camera Poses", show_label=False, visible=False) + basic_camera_motion_des = gr.Markdown(f"\n 1. Click one of the basic camera poses, such as `Pan Up`; \ + \n 2. Slide the `Motion speed` to get a speed value. The large the value, the fast the camera motion; \ + \n 3. Click `Visualize Camera and Proceed` to visualize the camera poses and go proceed; \ + \n 4. Click `Reset Camera` to reset the camera poses (If needed). ", + show_label=False, visible=False) + + + # step2.2 - camera motion control - provided complex + complex_camera_motion = gr.Markdown("---\n### Provided Complex Camera Poses", show_label=False, visible=False) + complex_camera_motion_des = gr.Markdown(f"\n 1. Click one of the complex camera poses, such as `Pose_1`; \ + \n 2. Click `Visualize Camera and Proceed` to visualize the camera poses and go proceed; \ + \n 3. Click `Reset Camera` to reset the camera poses (If needed). ", + show_label=False, visible=False) + + # step2.3 - camera motion control - custom + custom_camera_motion = gr.Markdown("---\n### Custom Camera Poses", show_label=False, visible=False) + custom_run_status = gr.Markdown(f"\n 1. Click two of the basic camera poses, such as `Pan Up` and `Pan Left`; \ + \n 2. Click `Customized Mode 1: First A then B` or `Customized Mode 1: First A then B` \ + \n - `Customized Mode 1: First A then B`: The camera first `Pan Up` and then `Pan Left`; \ + \n - `Customized Mode 2: Both A and B`: The camera move towards the upper left corner; \ + \n 3. Slide the `Motion speed` to get a speed value. The large the value, the fast the camera motion; \ + \n 4. Click `Visualize Camera and Proceed` to visualize the camera poses and go proceed; \ + \n 5. Click `Reset Camera` to reset the camera poses (If needed). ", + show_label=False, visible=False) + + gr.HighlightedText(value=[("",""), ("1. Select two of the basic camera poses; 2. Select Customized Mode 1 OR Customized Mode 2. 3. Visualized Camera to show the customized camera poses", "Normal")], + color_map={"Normal": "green", "Error": "red", "Clear clicks": "gray", "Add mask": "green", "Remove mask": "red"}, visible=False) + + with gr.Row(): + U = gr.Button(value="Pan Up", visible=False) + D = gr.Button(value="Pan Down", visible=False) + L = gr.Button(value="Pan Left", visible=False) + R = gr.Button(value="Pan Right", visible=False) + with gr.Row(): + I = gr.Button(value="Zoom In", visible=False) + O = gr.Button(value="Zoom Out", visible=False) + ACW = gr.Button(value="ACW", visible=False) + CW = gr.Button(value="CW", visible=False) + + with gr.Row(): + combine1 = gr.Button(value="Customized Mode 1: First A then B", visible=False) + combine2 = gr.Button(value="Customized Mode 2: Both A and B", visible=False) + + with gr.Row(): + speed = gr.Slider(minimum=0, maximum=2, step=0.2, label="Motion Speed", value=1.0, visible=False) + + with gr.Row(): + Pose_1 = gr.Button(value="Pose_1", visible=False) + Pose_2 = gr.Button(value="Pose_2", visible=False) + Pose_3 = gr.Button(value="Pose_3", visible=False) + Pose_4 = gr.Button(value="Pose_4", visible=False) + with gr.Row(): + Pose_5 = gr.Button(value="Pose_5", visible=False) + Pose_6 = gr.Button(value="Pose_6", visible=False) + Pose_7 = gr.Button(value="Pose_7", visible=False) + Pose_8 = gr.Button(value="Pose_8", visible=False) + + with gr.Row(): + camera_args = gr.Textbox(value="Camera Type", label="Camera Type", visible=False) + with gr.Row(): + camera_vis= gr.Button(value="Visualize Camera and Proceed", visible=False) + camera_reset = gr.Button(value="Reset Camera", visible=False) + with gr.Column(): + vis_camera = gr.Plot(fig, label='Camera Poses', visible=False) + + # step2 - object motion control + step2_object_motion = gr.Markdown("---\n## Step 2/3: Select a Provided Trajectory of Draw Yourself", show_label=False, visible=False) + step2_object_motion_des = gr.Markdown(f"\n - {OBJECT_MOTION_MODE[0]}: We provide some example trajectories. You can select one of them directly. \ + \n - {OBJECT_MOTION_MODE[1]}: Draw the trajectory yourself. \ + \n - Click `Proceed` to go into next step", + show_label=False, visible=False) + + object_mode = gr.Radio(choices=OBJECT_MOTION_MODE, value=OBJECT_MOTION_MODE[0], label="Motion Control Mode", interactive=True, visible=False) + object_info = gr.Button(value="Proceed", visible=False) + + with gr.Row(): + with gr.Column(): + # step2.1 - object motion control - provided + provided_traj = gr.Markdown("---\n### Provided Trajectory", show_label=False, visible=False) + provided_traj_des = gr.Markdown(f"\n 1. Click one of the provided trajectories, such as `horizon_1`; \ + \n 2. Click `Visualize Trajectory and Proceed` to visualize the trajectory and go proceed; \ + \n 3. Click `Reset Trajectory` to reset the trajectory (If needed). ", + show_label=False, visible=False) + + # step2.2 - object motion control - draw yourself + draw_traj = gr.Markdown("---\n### Draw Yourself", show_label=False, visible=False) + draw_run_status = gr.Markdown(f"\n 1. Click the `Canvas` in the right to draw the trajectory. **Note that You have to click the canva many times. For time saving, \ + the click point will not appear in the canvas but its coordinates will be written in `Points of Trajectory`**; \ + \n 2. Click `Visualize Trajectory and Proceed` to visualize the trajectory and go proceed; \ + \n 3. Click `Reset Trajectory` to reset the trajectory (If needed). ", + show_label=False, visible=False) + + with gr.Row(): + traj_1 = gr.Button(value="horizon_1", visible=False) + traj_2 = gr.Button(value="swaying_1", visible=False) + traj_3 = gr.Button(value="swaying_2", visible=False) + traj_4 = gr.Button(value="swaying_3", visible=False) + with gr.Row(): + traj_5 = gr.Button(value="curve_1", visible=False) + traj_6 = gr.Button(value="curve_2", visible=False) + traj_7 = gr.Button(value="curve_3", visible=False) + traj_8 = gr.Button(value="curve_4", visible=False) + + traj_args = gr.Textbox(value="", label="Points of Trajectory", visible=False) + with gr.Row(): + traj_vis = gr.Button(value="Visualize Trajectory and Proceed", visible=False) + traj_reset = gr.Button(value="Reset Trajectory", visible=False) + traj_droplast = gr.Button(value="Drop Last Point", visible=False) + + with gr.Column(): + traj_input = gr.Image("assets/traj_layout.png", tool='sketch', source="canvas", + width=256, height=256, + label="Canvas for Drawing", visible=False) + vis_traj = gr.Video(value=None, label="Trajectory", visible=False, width=256, height=256) + + + + # step3 - Add prompt and Generate videos + with gr.Row(): + with gr.Column(): + step3_prompt_generate = gr.Markdown("---\n## Step 3/3: Add prompt and Generate videos", show_label=False, visible=False) + prompt = gr.Textbox(value="a dog sitting on grass", label="Prompt", interactive=True, visible=False) + n_samples = gr.Number(value=3, precision=0, interactive=True, label="n_samples", visible=False) + seed = gr.Number(value=1234, precision=0, interactive=True, label="Seed", visible=False) + start = gr.Button(value="Start generation !", visible=False) + with gr.Column(): + gen_video = gr.Video(value=None, label="Generate Video", visible=False) + + mode_info.click( + fn=visualized_step2, + inputs=[infer_mode], + outputs=[step2_camera_motion, + step2_camera_motion_des, + camera_mode, + camera_info, + + basic_camera_motion, + basic_camera_motion_des, + custom_camera_motion, + custom_run_status, + complex_camera_motion, + complex_camera_motion_des, + U, D, L, R, + I, O, ACW, CW, + combine1, combine2, + speed, + Pose_1, Pose_2, Pose_3, Pose_4, + Pose_5, Pose_6, Pose_7, Pose_8, + camera_args, + camera_reset, camera_vis, + vis_camera, + + step2_object_motion, + step2_object_motion_des, + object_mode, + object_info, + + provided_traj, + provided_traj_des, + draw_traj, + draw_run_status, + traj_1, traj_2, traj_3, traj_4, + traj_5, traj_6, traj_7, traj_8, + traj_args, + traj_droplast, traj_reset, + traj_vis, + traj_input, vis_traj, + + step2_camera_object_motion, + step2_camera_object_motion_des, + + step3_prompt_generate, prompt, n_samples, seed, start, gen_video, + + ], + ) + + camera_info.click( + fn=visualized_camera_poses, + inputs=[camera_mode], + outputs=[basic_camera_motion, + basic_camera_motion_des, + custom_camera_motion, + custom_run_status, + complex_camera_motion, + complex_camera_motion_des, + U, D, L, R, + I, O, ACW, CW, + combine1, combine2, + speed, + Pose_1, Pose_2, Pose_3, Pose_4, + Pose_5, Pose_6, Pose_7, Pose_8, + camera_args, + camera_reset, camera_vis, + vis_camera, + step3_prompt_generate, prompt, n_samples, seed, start, gen_video], + ) + + object_info.click( + fn=visualized_traj_poses, + inputs=[object_mode], + outputs=[provided_traj, + provided_traj_des, + draw_traj, + draw_run_status, + traj_1, traj_2, traj_3, traj_4, + traj_5, traj_6, traj_7, traj_8, + traj_args, + traj_droplast, traj_reset, + traj_vis, + traj_input, vis_traj, + step3_prompt_generate, prompt, n_samples, seed, start, gen_video,], + ) + + + U.click(fn=add_camera_motion, inputs=[U, camera_mode], outputs=camera_args) + D.click(fn=add_camera_motion, inputs=[D, camera_mode], outputs=camera_args) + L.click(fn=add_camera_motion, inputs=[L, camera_mode], outputs=camera_args) + R.click(fn=add_camera_motion, inputs=[R, camera_mode], outputs=camera_args) + I.click(fn=add_camera_motion, inputs=[I, camera_mode], outputs=camera_args) + O.click(fn=add_camera_motion, inputs=[O, camera_mode], outputs=camera_args) + ACW.click(fn=add_camera_motion, inputs=[ACW, camera_mode], outputs=camera_args) + CW.click(fn=add_camera_motion, inputs=[CW, camera_mode], outputs=camera_args) + speed.change(fn=change_camera_speed, inputs=speed, outputs=camera_args) + camera_reset.click(fn=reset_camera, inputs=None, outputs=[camera_args]) + + combine1.click(fn=change_camera_mode, inputs=[combine1, camera_mode], outputs=camera_args) + combine2.click(fn=change_camera_mode, inputs=[combine2, camera_mode], outputs=camera_args) + + camera_vis.click(fn=fn_vis_camera, inputs=[infer_mode], outputs=[vis_camera, object_mode, object_info, step3_prompt_generate, prompt, n_samples, seed, start, gen_video]) + + Pose_1.click(fn=add_complex_camera_motion, inputs=Pose_1, outputs=camera_args) + Pose_2.click(fn=add_complex_camera_motion, inputs=Pose_2, outputs=camera_args) + Pose_3.click(fn=add_complex_camera_motion, inputs=Pose_3, outputs=camera_args) + Pose_4.click(fn=add_complex_camera_motion, inputs=Pose_4, outputs=camera_args) + Pose_5.click(fn=add_complex_camera_motion, inputs=Pose_5, outputs=camera_args) + Pose_6.click(fn=add_complex_camera_motion, inputs=Pose_6, outputs=camera_args) + Pose_7.click(fn=add_complex_camera_motion, inputs=Pose_7, outputs=camera_args) + Pose_8.click(fn=add_complex_camera_motion, inputs=Pose_8, outputs=camera_args) + + traj_1.click(fn=add_provided_traj, inputs=traj_1, outputs=traj_args) + traj_2.click(fn=add_provided_traj, inputs=traj_2, outputs=traj_args) + traj_3.click(fn=add_provided_traj, inputs=traj_3, outputs=traj_args) + traj_4.click(fn=add_provided_traj, inputs=traj_4, outputs=traj_args) + traj_5.click(fn=add_provided_traj, inputs=traj_5, outputs=traj_args) + traj_6.click(fn=add_provided_traj, inputs=traj_6, outputs=traj_args) + traj_7.click(fn=add_provided_traj, inputs=traj_7, outputs=traj_args) + traj_8.click(fn=add_provided_traj, inputs=traj_8, outputs=traj_args) + + traj_vis.click(fn=fn_vis_traj, inputs=None, outputs=[vis_traj, step3_prompt_generate, prompt, n_samples, seed, start, gen_video]) + traj_input.select(fn=add_traj_point, inputs=None, outputs=traj_args) + traj_droplast.click(fn=fn_traj_droplast, inputs=None, outputs=traj_args) + traj_reset.click(fn=fn_traj_reset, inputs=None, outputs=traj_args) + + + start.click(fn=model_run, inputs=[prompt, infer_mode, seed, n_samples], outputs=gen_video) + + gr.Markdown(article) + + # demo.launch(server_name='0.0.0.0', share=False, server_port=args.port) + demo.queue(concurrency_count=1, max_size=10) + demo.launch() + + + +if __name__=="__main__": + parser = argparse.ArgumentParser() + parser.add_argument("--port", type=int, default=12345) + args = parser.parse_args() + + main(args) diff --git a/assets/traj_layout.png b/assets/traj_layout.png new file mode 100644 index 0000000000000000000000000000000000000000..17761ff18b17eb6e31ec0ef0f8d9eb9acd9c6235 Binary files /dev/null and b/assets/traj_layout.png differ diff --git a/configs/inference/config_both.yaml b/configs/inference/config_both.yaml new file mode 100644 index 0000000000000000000000000000000000000000..47fef6c961c4ca3005f7e6b666e68e4dca742d01 --- /dev/null +++ b/configs/inference/config_both.yaml @@ -0,0 +1,104 @@ +model: + base_learning_rate: 0.0001 + scale_lr: false + target: motionctrl.motionctrl.MotionCtrl + params: + # param for object motion control + omcm_config: + pretrained: ~ + target: lvdm.modules.encoders.adapter.Adapter + params: + channels: + - 320 + - 640 + - 1280 + - 1280 + nums_rb: 2 + cin: 128 + sk: true + use_conv: false + + linear_start: 0.00085 + linear_end: 0.012 + num_timesteps_cond: 1 + log_every_t: 200 + timesteps: 1000 + first_stage_key: video + cond_stage_key: caption + cond_stage_trainable: false + conditioning_key: crossattn + image_size: + - 32 + - 32 + channels: 4 + scale_by_std: false + scale_factor: 0.18215 + use_ema: false + uncond_prob: 0.1 + uncond_type: empty_seq + empty_params_only: true + scheduler_config: + target: utils.lr_scheduler.LambdaLRScheduler + interval: step + frequency: 100 + params: + start_step: 0 + final_decay_ratio: 0.01 + decay_steps: 20000 + unet_config: + target: lvdm.modules.networks.openaimodel3d_next.UNetModel + params: + in_channels: 4 + out_channels: 4 + model_channels: 320 + attention_resolutions: + - 4 + - 2 + - 1 + num_res_blocks: 2 + channel_mult: + - 1 + - 2 + - 4 + - 4 + num_head_channels: 64 + transformer_depth: 1 + context_dim: 1024 + use_linear: true + use_checkpoint: true + temporal_conv: true + temporal_attention: true + temporal_selfatt_only: true + use_relative_position: false + use_causal_attention: false + temporal_length: 16 + use_image_dataset: false + addition_attention: true + first_stage_config: + target: lvdm.models.autoencoder.AutoencoderKL + params: + embed_dim: 4 + monitor: val/rec_loss + ddconfig: + double_z: true + z_channels: 4 + resolution: 256 + in_channels: 3 + out_ch: 3 + ch: 128 + ch_mult: + - 1 + - 2 + - 4 + - 4 + num_res_blocks: 2 + attn_resolutions: [] + dropout: 0.0 + lossconfig: + target: torch.nn.Identity + cond_stage_config: + target: lvdm.modules.encoders.condition2.FrozenOpenCLIPEmbedder + params: + freeze: true + layer: penultimate + \ No newline at end of file diff --git a/configs/inference/run.sh b/configs/inference/run.sh new file mode 100644 index 0000000000000000000000000000000000000000..366da483e75a9763489ff2b97518fa61f4101888 --- /dev/null +++ b/configs/inference/run.sh @@ -0,0 +1,52 @@ + +config="configs/inference/config_both.yaml" +ckpt='./checkpoints/motionctrl.pth' + +condtype='both' +condtype='object_motion' +condtype='camera_motion' + +cond_dir="examples/" + +res_dir="./outputs/" +if [ ! -d $res_dir ]; then + mkdir -p $res_dir +fi + +save_dir=$res_dir/$condtype'_seed'$seed + +use_ddp=0 + +if [ $use_ddp == 0 ]; then +CUDA_VISIBLE_DEVICES=7 python 'main/evaluation/motionctrl_inference.py' \ +--seed 1234 \ +--ckpt_path $ckpt \ +--base $config \ +--savedir $save_dir \ +--n_samples 5 \ +--bs 1 --height 256 --width 256 \ +--unconditional_guidance_scale 7.5 \ +--ddim_steps 50 \ +--ddim_eta 1.0 \ +--condtype $condtype \ +--cond_dir $cond_dir \ +# --save_imgs +fi + +if [ $use_ddp == 1 ]; then +python3 -m torch.distributed.launch \ +--nproc_per_node=3 --nnodes=1 --master_port=23466 \ +main/evaluation/ddp_wrapper.py \ +--module 'inference' \ +--seed 2000 \ +--ckpt_path $ckpt \ +--base $config \ +--savedir $res_dir/$name \ +--n_samples 3 \ +--bs 1 --height 256 --width 256 \ +--unconditional_guidance_scale 7.5 \ +--ddim_steps 50 \ +--ddim_eta 1.0 \ +--condtype $condtype \ +--cond_dir $cond_dir +fi \ No newline at end of file diff --git a/examples/camera_poses/test_camera_018f7907401f2fef.json b/examples/camera_poses/test_camera_018f7907401f2fef.json new file mode 100644 index 0000000000000000000000000000000000000000..c9cc8fa2c3ca1200ca75b2e3a4a381cf75d8f157 --- /dev/null +++ b/examples/camera_poses/test_camera_018f7907401f2fef.json @@ -0,0 +1 @@ +[[1.0, -4.493872218791495e-10, 5.58983348497577e-09, 1.9967236752904682e-09, -4.493872218791495e-10, 1.0, -6.144247333139674e-10, 1.0815730533408896e-09, 5.58983348497577e-09, -6.144247333139674e-10, 1.0, -7.984015226725205e-09], [0.9982863664627075, -0.0024742060340940952, 0.05846544727683067, -0.024547122418880463, 0.002410230925306678, 0.9999964237213135, 0.0011647245846688747, -0.003784072818234563, -0.05846811458468437, -0.0010218139505013824, 0.9982887506484985, -0.09103696048259735], [0.9933298230171204, -0.006303737405687571, 0.11513543128967285, -0.053876250982284546, 0.00586089538410306, 0.9999741315841675, 0.004184383898973465, -0.006566310301423073, -0.115158811211586, -0.0034816779661923647, 0.9933409690856934, -0.18525512516498566], [0.9849286675453186, -0.013619760051369667, 0.17242403328418732, -0.08322551101446152, 0.01256392989307642, 0.9998950958251953, 0.0072133541107177734, -0.004579910542815924, -0.17250417172908783, -0.004938316997140646, 0.9849964380264282, -0.28701746463775635], [0.9731453657150269, -0.022617166861891747, 0.2290775030851364, -0.11655563861131668, 0.02060025744140148, 0.9997251629829407, 0.011192308738827705, -0.0017426757840439677, -0.2292676568031311, -0.006172688212245703, 0.9733438491821289, -0.37736839056015015], [0.9582399725914001, -0.03294993191957474, 0.2840607464313507, -0.15743066370487213, 0.030182993039488792, 0.9994447827339172, 0.014113469049334526, -0.002769832033663988, -0.28436803817749023, -0.004950287751853466, 0.9587023854255676, -0.46959081292152405], [0.940129816532135, -0.03991429880261421, 0.3384712040424347, -0.22889098525047302, 0.03725311905145645, 0.9992027282714844, 0.01435780432075262, -0.0028311305213719606, -0.3387744128704071, -0.0008890923927538097, 0.9408671855926514, -0.5631460547447205], [0.9222924709320068, -0.044258520007133484, 0.38395029306411743, -0.2986142039299011, 0.04110203683376312, 0.9990199208259583, 0.01642671786248684, 0.0013055746676400304, -0.38430097699165344, 0.000630900904070586, 0.9232076406478882, -0.6414245367050171], [0.9061535000801086, -0.04851173609495163, 0.4201577305793762, -0.3483412563800812, 0.04521748423576355, 0.9988185167312622, 0.017803886905312538, 0.0010280977003276348, -0.4205249547958374, 0.0028654206544160843, 0.907276451587677, -0.7144853472709656], [0.8919307589530945, -0.05171844735741615, 0.4492044746875763, -0.37905213236808777, 0.04818608984351158, 0.9986518621444702, 0.019300933927297592, 0.00036871168413199484, -0.44959715008735657, 0.004430312197655439, 0.8932204246520996, -0.7976372241973877], [0.8792291879653931, -0.05425972864031792, 0.47329893708229065, -0.39671003818511963, 0.05076585337519646, 0.998507022857666, 0.02016463316977024, 0.001104982104152441, -0.4736863970756531, 0.00629808846861124, 0.8806710243225098, -0.8874085545539856], [0.8659296035766602, -0.0567130371928215, 0.49694016575813293, -0.4097800552845001, 0.05366959795355797, 0.9983500838279724, 0.020415671169757843, 0.0009228077251464128, -0.497278094291687, 0.008992047980427742, 0.8675445914268494, -0.9762357473373413], [0.8503361940383911, -0.055699657648801804, 0.5232837200164795, -0.44268566370010376, 0.054582174867391586, 0.9983546733856201, 0.01757136546075344, 0.005412018392235041, -0.5234014391899109, 0.013620397076010704, 0.8519773483276367, -1.069865107536316], [0.836037814617157, -0.05214058235287666, 0.5461887717247009, -0.4671085774898529, 0.05177384987473488, 0.9985294938087463, 0.01607322134077549, 0.008980141952633858, -0.5462236404418945, 0.014840473420917988, 0.8375079035758972, -1.1569048166275024], [0.82603919506073, -0.04987695440649986, 0.5614013671875, -0.4677649438381195, 0.05124447122216225, 0.9985973834991455, 0.013318539597094059, 0.012170637026429176, -0.5612781643867493, 0.017767081037163734, 0.8274364471435547, -1.2651430368423462], [0.8179472088813782, -0.0496118925511837, 0.573150098323822, -0.45822662115097046, 0.052784956991672516, 0.9985441565513611, 0.011104168370366096, 0.018991567194461823, -0.5728666186332703, 0.0211710836738348, 0.8193751573562622, -1.3895009756088257]] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_088b93f15ca8745d.json b/examples/camera_poses/test_camera_088b93f15ca8745d.json new file mode 100644 index 0000000000000000000000000000000000000000..eeabf8894882e430e672f1cfeb5003ca419da7dc --- /dev/null +++ b/examples/camera_poses/test_camera_088b93f15ca8745d.json @@ -0,0 +1 @@ +[[0.9999999403953552, 3.8618797049139175e-10, -1.3441345814158012e-08, 1.3928219289027766e-07, 3.8618797049139175e-10, 1.0, -4.134579345560496e-10, -6.074658998045379e-09, -1.3441345814158012e-08, -4.134579345560496e-10, 1.0, 7.038884319854333e-08], [0.9994913339614868, 0.003077245783060789, -0.031741149723529816, 0.08338673412799835, -0.0030815028585493565, 0.999995231628418, -8.520588744431734e-05, 0.006532138213515282, 0.0317407064139843, 0.00018297435599379241, 0.9994961619377136, -0.02256060019135475], [0.9979938268661499, 0.0051255361177027225, -0.06310292333364487, 0.18344485759735107, -0.005117486696690321, 0.9999868869781494, 0.00028916727751493454, 0.018134046345949173, 0.06310353428125381, 3.434090831433423e-05, 0.9980069994926453, -0.030579563230276108], [0.9954646825790405, 0.00820203311741352, -0.0947771891951561, 0.29663264751434326, -0.00811922550201416, 0.9999662041664124, 0.0012593322899192572, 0.02404301054775715, 0.09478426724672318, -0.0004841022891923785, 0.9954977035522461, -0.02678978443145752], [0.9913660883903503, 0.012001598253846169, -0.13057230412960052, 0.4076530337333679, -0.011968829669058323, 0.999927818775177, 0.0010357286082580686, 0.024977533146739006, 0.1305752843618393, 0.0005360084469430149, 0.9914382100105286, -0.010779343545436859], [0.985666811466217, 0.017323914915323257, -0.16781197488307953, 0.509911060333252, -0.017399737611413002, 0.9998481273651123, 0.0010186078725382686, 0.023117201402783394, 0.16780413687229156, 0.0019158748909831047, 0.9858185052871704, 0.018053216859698296], [0.9784473180770874, 0.022585421800613403, -0.20525763928890228, 0.5957884192466736, -0.022850200533866882, 0.9997382760047913, 0.0010805513011291623, 0.020451901480555534, 0.2052282989025116, 0.003632916137576103, 0.9787073731422424, 0.03460140898823738], [0.9711515307426453, 0.026846906170248985, -0.23694702982902527, 0.6832671165466309, -0.02745947800576687, 0.999622642993927, 0.0007151798927225173, 0.012211678549647331, 0.23687675595283508, 0.005811895243823528, 0.971522331237793, 0.03236595541238785], [0.9641746878623962, 0.030338184908032417, -0.26352745294570923, 0.7764986157417297, -0.031404945999383926, 0.9995067715644836, 0.0001645474840188399, 0.0011497576488181949, 0.26340243220329285, 0.008117412216961384, 0.964651882648468, 0.022656364366412163], [0.9573631882667542, 0.0335896760225296, -0.2869274914264679, 0.8815275430679321, -0.03532479330897331, 0.9993755221366882, -0.0008711823611520231, -0.003618708113208413, 0.2867189943790436, 0.010969695635139942, 0.9579519033432007, 0.005283573176711798], [0.9507063627243042, 0.036557890474796295, -0.3079299330711365, 0.9931321740150452, -0.03846294432878494, 0.9992600679397583, -0.00011733790597645566, 0.0018704120302572846, 0.30769774317741394, 0.01195544097572565, 0.9514090418815613, -0.035360634326934814], [0.9448517560958862, 0.039408694952726364, -0.3251185715198517, 1.1025006771087646, -0.041503626853227615, 0.9991382360458374, 0.0004919985658489168, 0.007425118237733841, 0.32485777139663696, 0.013028733432292938, 0.9456731081008911, -0.09869624674320221], [0.940796971321106, 0.04081147164106369, -0.33650481700897217, 1.1961394548416138, -0.0429220013320446, 0.9990777373313904, 0.0011677180882543325, 0.019955899566411972, 0.336242139339447, 0.013344875536859035, 0.9416810274124146, -0.16835527122020721], [0.9376427531242371, 0.04111124947667122, -0.3451607823371887, 1.2392503023147583, -0.043144747614860535, 0.9990671873092651, 0.0017920633545145392, 0.03982722759246826, 0.34491249918937683, 0.013211555778980255, 0.938541829586029, -0.24618202447891235], [0.9353355765342712, 0.04122937470674515, -0.3513509929180145, 1.285768747329712, -0.043183211237192154, 0.9990646243095398, 0.0022769556380808353, 0.06841164082288742, 0.3511161506175995, 0.01304274145513773, 0.936241090297699, -0.3213619291782379], [0.9342393279075623, 0.041213057935237885, -0.3542574644088745, 1.3363462686538696, -0.04236872121691704, 0.9990919232368469, 0.0044970144517719746, 0.08925694227218628, 0.35412102937698364, 0.010808154009282589, 0.9351370930671692, -0.40201041102409363]] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_1424acd0007d40b5.json b/examples/camera_poses/test_camera_1424acd0007d40b5.json new file mode 100644 index 0000000000000000000000000000000000000000..02b37fd5a70f4db9b156257145734dc8f6b1499e --- /dev/null +++ b/examples/camera_poses/test_camera_1424acd0007d40b5.json @@ -0,0 +1 @@ +[[1.0, 9.44418099280142e-10, 3.889182664806867e-08, 6.214055492392845e-09, 9.44418099280142e-10, 1.0, -1.0644604121756718e-11, -7.621465680784922e-10, 3.889182664806867e-08, -1.0644604121756718e-11, 1.0, -2.7145965475483536e-08], [0.9873979091644287, -0.007892023772001266, 0.15806053578853607, 0.4749181270599365, 0.008024877868592739, 0.9999678134918213, -0.00020230526570230722, 0.1585356593132019, -0.15805381536483765, 0.0014681711327284575, 0.9874294400215149, -0.2091633826494217], [0.9708925485610962, -0.011486345902085304, 0.23923994600772858, 0.8120080828666687, 0.012198254466056824, 0.9999244809150696, -0.0014952132478356361, 0.2486257702112198, -0.23920467495918274, 0.004370000213384628, 0.9709593057632446, -0.5957822799682617], [0.9619541168212891, -0.013188007287681103, 0.2728927433490753, 1.1486873626708984, 0.014017474837601185, 0.9999011754989624, -0.001090032048523426, 0.3114692270755768, -0.2728513777256012, 0.0048738280311226845, 0.962043821811676, -1.0323039293289185], [0.9586812257766724, -0.013936692848801613, 0.284140944480896, 1.5948307514190674, 0.014867136254906654, 0.9998888373374939, -0.0011181083973497152, 0.36000898480415344, -0.2840937674045563, 0.0052962712943553925, 0.958781898021698, -1.4377187490463257], [0.9583359360694885, -0.011928150430321693, 0.28539448976516724, 2.002793788909912, 0.014221147634088993, 0.9998810887336731, -0.0059633455239236355, 0.35464340448379517, -0.28528934717178345, 0.009773525409400463, 0.9583916068077087, -1.8953297138214111], [0.9584393501281738, -0.010862396098673344, 0.28508952260017395, 2.3351645469665527, 0.012857729569077492, 0.9999041557312012, -0.005128204356878996, 0.38934090733528137, -0.28500640392303467, 0.008580676279962063, 0.9584871530532837, -2.4214961528778076], [0.9587277173995972, -0.009760312736034393, 0.28415825963020325, 2.6858017444610596, 0.012186127714812756, 0.9999027848243713, -0.0067702098749578, 0.4173329174518585, -0.28406453132629395, 0.009953574277460575, 0.9587535262107849, -3.030754327774048], [0.9589635729789734, -0.0070899901911616325, 0.2834406793117523, 2.8917219638824463, 0.010482418350875378, 0.9998904466629028, -0.01045384630560875, 0.4001043438911438, -0.28333547711372375, 0.012996001169085503, 0.9589328169822693, -3.6960957050323486], [0.9590328931808472, -0.005921780597418547, 0.2832328677177429, 3.034579038619995, 0.00947173498570919, 0.9998928308486938, -0.01116593275219202, 0.4193899631500244, -0.2831363379955292, 0.013391205109655857, 0.9589861631393433, -4.384733200073242], [0.9593284726142883, -0.004661113955080509, 0.28225383162498474, 3.2042288780212402, 0.0076906089670956135, 0.9999240636825562, -0.009626304730772972, 0.4752484858036041, -0.28218749165534973, 0.011405492201447487, 0.9592914581298828, -5.098723411560059], [0.9591755867004395, -0.0035665074829012156, 0.2827887237071991, 3.263953924179077, 0.0062992447055876255, 0.9999418258666992, -0.008754877373576164, 0.4543868899345398, -0.282740980386734, 0.010178821161389351, 0.95914226770401, -5.7807512283325195], [0.9591742753982544, -0.003413048107177019, 0.2827949523925781, 3.3116462230682373, 0.003146615345031023, 0.9999940991401672, 0.0013963348465040326, 0.4299861788749695, -0.28279799222946167, -0.0004494813329074532, 0.9591793417930603, -6.478931903839111], [0.9585762619972229, -0.002857929328456521, 0.28482162952423096, 3.4120190143585205, -0.008201238699257374, 0.9992581605911255, 0.0376281812787056, 0.21596357226371765, -0.2847178280353546, -0.03840537369251251, 0.957841694355011, -7.178638935089111], [0.9572952389717102, -0.002719884505495429, 0.28909942507743835, 3.4662365913391113, -0.030634215101599693, 0.9933721423149109, 0.11078491061925888, -0.29767531156539917, -0.2874845862388611, -0.11491020023822784, 0.9508671164512634, -7.794575214385986], [0.9545961618423462, -0.005338957067579031, 0.29785510897636414, 3.5083835124969482, -0.06037351116538048, 0.9756243824958801, 0.2109788954257965, -1.0968165397644043, -0.29172107577323914, -0.21938219666481018, 0.9310050010681152, -8.306528091430664]] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_D.json b/examples/camera_poses/test_camera_D.json new file mode 100644 index 0000000000000000000000000000000000000000..0fa6462cba65992ec66f014933ca7c3e1ac9f9a8 --- /dev/null +++ b/examples/camera_poses/test_camera_D.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.2, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.28750000000000003, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.37500000000000006, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.4625000000000001, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.55, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.6375000000000002, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.7250000000000001, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.8125000000000002, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.9000000000000001, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -0.9875000000000003, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -1.0750000000000002, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -1.1625000000000003, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -1.2500000000000002, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -1.3375000000000001, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -1.4250000000000003, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + -1.5125000000000004, + 0.0, + 0.0, + 1.0, + 0.0 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_I.json b/examples/camera_poses/test_camera_I.json new file mode 100644 index 0000000000000000000000000000000000000000..a44cc11f8bad536d2feb86a56f3e45441f4d12b9 --- /dev/null +++ b/examples/camera_poses/test_camera_I.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.2 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.28750000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.37500000000000006 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.4625000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.55 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.6375000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.7250000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.8125000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.9000000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.9875000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.0750000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.1625000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.2500000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.3375000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.4250000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.5125000000000004 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_I_0.2x.json b/examples/camera_poses/test_camera_I_0.2x.json new file mode 100644 index 0000000000000000000000000000000000000000..c4c72597b15c09450618be858523e77f5c16d2b2 --- /dev/null +++ b/examples/camera_poses/test_camera_I_0.2x.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.022500000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.045000000000000005 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.0675 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.09000000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.11250000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.135 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.15750000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.18000000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.2025 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.22500000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.24750000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.27 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.29250000000000004 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.31500000000000006 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.3375 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_I_0.4x.json b/examples/camera_poses/test_camera_I_0.4x.json new file mode 100644 index 0000000000000000000000000000000000000000..2359910f99fc32b0d54cfda17ad864afcadbf5c7 --- /dev/null +++ b/examples/camera_poses/test_camera_I_0.4x.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.045000000000000005 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.09000000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.135 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.18000000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.22500000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.27 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.31500000000000006 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.36000000000000004 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.405 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.45000000000000007 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.49500000000000005 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.54 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.5850000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.6300000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.675 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_I_1.0x.json b/examples/camera_poses/test_camera_I_1.0x.json new file mode 100644 index 0000000000000000000000000000000000000000..46e3f75ee8ea4f0b204f27690dbccb61e814f7cd --- /dev/null +++ b/examples/camera_poses/test_camera_I_1.0x.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.1125 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.225 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.3375 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.45 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.5625 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.675 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.7875 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.9 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.0125 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.125 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.2375 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.35 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.4625000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.575 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.6875 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_I_2.0x.json b/examples/camera_poses/test_camera_I_2.0x.json new file mode 100644 index 0000000000000000000000000000000000000000..81207f6ef53151f1e01c771d0e13599d0183968b --- /dev/null +++ b/examples/camera_poses/test_camera_I_2.0x.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.225 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.45 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.675 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -0.9 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.125 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.35 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.575 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -1.8 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -2.025 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -2.25 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -2.475 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -2.7 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -2.9250000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -3.15 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + -3.375 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_L.json b/examples/camera_poses/test_camera_L.json new file mode 100644 index 0000000000000000000000000000000000000000..992b371f2e14ec52574f6777c78c89ecd23119b4 --- /dev/null +++ b/examples/camera_poses/test_camera_L.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.2, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.28750000000000003, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.37500000000000006, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.4625000000000001, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.55, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.6375000000000002, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.7250000000000001, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.8125000000000002, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.9000000000000001, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.9875000000000003, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 1.0750000000000002, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 1.1625000000000003, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 1.2500000000000002, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 1.3375000000000001, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 1.4250000000000003, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 1.5125000000000004, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_O.json b/examples/camera_poses/test_camera_O.json new file mode 100644 index 0000000000000000000000000000000000000000..b0bacc0cbf544e1fc4ee96047bd18e6cc8da0e9c --- /dev/null +++ b/examples/camera_poses/test_camera_O.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.2 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.28750000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.37500000000000006 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.4625000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.55 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.6375000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.7250000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.8125000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.9000000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.9875000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.0750000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.1625000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.2500000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.3375000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.4250000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.5125000000000004 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_O_0.2x.json b/examples/camera_poses/test_camera_O_0.2x.json new file mode 100644 index 0000000000000000000000000000000000000000..689535fb82ffc79c2ef0123a42a9b3c1b2dca5a3 --- /dev/null +++ b/examples/camera_poses/test_camera_O_0.2x.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.022500000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.045000000000000005 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0675 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.09000000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.11250000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.135 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.15750000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.18000000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.2025 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.22500000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.24750000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.27 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.29250000000000004 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.31500000000000006 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.3375 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_O_0.4x.json b/examples/camera_poses/test_camera_O_0.4x.json new file mode 100644 index 0000000000000000000000000000000000000000..01b6ede93af2e2daecbc325d029046195bed7ac0 --- /dev/null +++ b/examples/camera_poses/test_camera_O_0.4x.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.045000000000000005 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.09000000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.135 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.18000000000000002 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.22500000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.27 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.31500000000000006 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.36000000000000004 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.405 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.45000000000000007 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.49500000000000005 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.54 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.5850000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.6300000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.675 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_O_1.0x.json b/examples/camera_poses/test_camera_O_1.0x.json new file mode 100644 index 0000000000000000000000000000000000000000..4b91d2bf8993e20c20b0d9dc17870e4854aeef9a --- /dev/null +++ b/examples/camera_poses/test_camera_O_1.0x.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.1125 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.225 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.3375 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.45 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.5625 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.675 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.7875 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.9 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.0125 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.125 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.2375 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.35 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.4625000000000001 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.575 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.6875 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_O_2.0x.json b/examples/camera_poses/test_camera_O_2.0x.json new file mode 100644 index 0000000000000000000000000000000000000000..5481cc64af88c55ccb3ba8e7ac897ad96fdd88c3 --- /dev/null +++ b/examples/camera_poses/test_camera_O_2.0x.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.225 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.45 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.675 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.9 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.125 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.35 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.575 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.8 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 2.025 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 2.25 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 2.475 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 2.7 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 2.9250000000000003 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 3.15 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 3.375 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_R.json b/examples/camera_poses/test_camera_R.json new file mode 100644 index 0000000000000000000000000000000000000000..93be7cff1ce3dc0f4663e6394a6bda23c78bba35 --- /dev/null +++ b/examples/camera_poses/test_camera_R.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + -0.2, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.28750000000000003, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.37500000000000006, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.4625000000000001, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.55, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.6375000000000002, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.7250000000000001, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.8125000000000002, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.9000000000000001, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -0.9875000000000003, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -1.0750000000000002, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -1.1625000000000003, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -1.2500000000000002, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -1.3375000000000001, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -1.4250000000000003, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + -1.5125000000000004, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_Round-RI-120.json b/examples/camera_poses/test_camera_Round-RI-120.json new file mode 100644 index 0000000000000000000000000000000000000000..1798456ed380de19d1711035fea74d57c33ebb4e --- /dev/null +++ b/examples/camera_poses/test_camera_Round-RI-120.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.8 + ], + [ + 0.9914448613738104, + 0.0, + 0.13052619222005157, + 0.13052619222005157, + 0.0, + 1.0, + 0.0, + 0.0, + -0.13052619222005157, + 0.0, + 0.9914448613738104, + 1.7914448613738103 + ], + [ + 0.9659258262890683, + 0.0, + 0.25881904510252074, + 0.25881904510252074, + 0.0, + 1.0, + 0.0, + 0.0, + -0.25881904510252074, + 0.0, + 0.9659258262890683, + 1.7659258262890685 + ], + [ + 0.9238795325112867, + 0.0, + 0.3826834323650898, + 0.3826834323650898, + 0.0, + 1.0, + 0.0, + 0.0, + -0.3826834323650898, + 0.0, + 0.9238795325112867, + 1.7238795325112868 + ], + [ + 0.8660254037844387, + 0.0, + 0.49999999999999994, + 0.49999999999999994, + 0.0, + 1.0, + 0.0, + 0.0, + -0.49999999999999994, + 0.0, + 0.8660254037844387, + 1.6660254037844386 + ], + [ + 0.7933533402912353, + 0.0, + 0.6087614290087205, + 0.6087614290087205, + 0.0, + 1.0, + 0.0, + 0.0, + -0.6087614290087205, + 0.0, + 0.7933533402912353, + 1.5933533402912352 + ], + [ + 0.7071067811865476, + 0.0, + 0.7071067811865476, + 0.7071067811865476, + 0.0, + 1.0, + 0.0, + 0.0, + -0.7071067811865476, + 0.0, + 0.7071067811865476, + 1.5071067811865477 + ], + [ + 0.6087614290087207, + 0.0, + 0.7933533402912352, + 0.7933533402912352, + 0.0, + 1.0, + 0.0, + 0.0, + -0.7933533402912352, + 0.0, + 0.6087614290087207, + 1.4087614290087207 + ], + [ + 0.5000000000000001, + 0.0, + 0.8660254037844386, + 0.8660254037844386, + 0.0, + 1.0, + 0.0, + 0.0, + -0.8660254037844386, + 0.0, + 0.5000000000000001, + 1.3000000000000003 + ], + [ + 0.38268343236508984, + 0.0, + 0.9238795325112867, + 0.9238795325112867, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9238795325112867, + 0.0, + 0.38268343236508984, + 1.1826834323650899 + ], + [ + 0.25881904510252096, + 0.0, + 0.9659258262890682, + 0.9659258262890682, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9659258262890682, + 0.0, + 0.25881904510252096, + 1.058819045102521 + ], + [ + 0.1305261922200517, + 0.0, + 0.9914448613738104, + 0.9914448613738104, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9914448613738104, + 0.0, + 0.1305261922200517, + 0.9305261922200517 + ], + [ + 6.123233995736766e-17, + 0.0, + 1.0, + 1.0, + 0.0, + 1.0, + 0.0, + 0.0, + -1.0, + 0.0, + 6.123233995736766e-17, + 0.8000000000000002 + ], + [ + -0.13052619222005138, + 0.0, + 0.9914448613738105, + 0.9914448613738105, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9914448613738105, + 0.0, + -0.13052619222005138, + 0.6694738077799487 + ], + [ + -0.25881904510252063, + 0.0, + 0.9659258262890683, + 0.9659258262890683, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9659258262890683, + 0.0, + -0.25881904510252063, + 0.5411809548974794 + ], + [ + -0.3826834323650895, + 0.0, + 0.9238795325112868, + 0.9238795325112868, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9238795325112868, + 0.0, + -0.3826834323650895, + 0.41731656763491054 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_Round-RI.json b/examples/camera_poses/test_camera_Round-RI.json new file mode 100644 index 0000000000000000000000000000000000000000..0c2c7a00d262c3f3363fa2c8a613a29377b14c25 --- /dev/null +++ b/examples/camera_poses/test_camera_Round-RI.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.7000000000000002 + ], + [ + 0.9807852804032304, + 0.0, + 0.19509032201612825, + 0.17558128981451543, + 0.0, + 1.0, + 0.0, + 0.0, + -0.19509032201612825, + 0.0, + 0.9807852804032304, + 1.6827067523629076 + ], + [ + 0.9238795325112867, + 0.0, + 0.3826834323650898, + 0.3444150891285808, + 0.0, + 1.0, + 0.0, + 0.0, + -0.3826834323650898, + 0.0, + 0.9238795325112867, + 1.631491579260158 + ], + [ + 0.8314696123025452, + 0.0, + 0.5555702330196022, + 0.500013209717642, + 0.0, + 1.0, + 0.0, + 0.0, + -0.5555702330196022, + 0.0, + 0.8314696123025452, + 1.5483226510722907 + ], + [ + 0.7071067811865476, + 0.0, + 0.7071067811865476, + 0.6363961030678928, + 0.0, + 1.0, + 0.0, + 0.0, + -0.7071067811865476, + 0.0, + 0.7071067811865476, + 1.436396103067893 + ], + [ + 0.5555702330196023, + 0.0, + 0.8314696123025452, + 0.7483226510722907, + 0.0, + 1.0, + 0.0, + 0.0, + -0.8314696123025452, + 0.0, + 0.5555702330196023, + 1.3000132097176422 + ], + [ + 0.38268343236508984, + 0.0, + 0.9238795325112867, + 0.831491579260158, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9238795325112867, + 0.0, + 0.38268343236508984, + 1.144415089128581 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.8827067523629074, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.9755812898145155 + ], + [ + 6.123233995736766e-17, + 0.0, + 1.0, + 0.9, + 0.0, + 1.0, + 0.0, + 0.0, + -1.0, + 0.0, + 6.123233995736766e-17, + 0.8 + ], + [ + -0.1950903220161282, + 0.0, + 0.9807852804032304, + 0.8827067523629074, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + -0.1950903220161282, + 0.6244187101854847 + ], + [ + -0.3826834323650897, + 0.0, + 0.9238795325112867, + 0.831491579260158, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9238795325112867, + 0.0, + -0.3826834323650897, + 0.4555849108714193 + ], + [ + -0.555570233019602, + 0.0, + 0.8314696123025453, + 0.7483226510722908, + 0.0, + 1.0, + 0.0, + 0.0, + -0.8314696123025453, + 0.0, + -0.555570233019602, + 0.2999867902823583 + ], + [ + -0.7071067811865475, + 0.0, + 0.7071067811865476, + 0.6363961030678928, + 0.0, + 1.0, + 0.0, + 0.0, + -0.7071067811865476, + 0.0, + -0.7071067811865475, + 0.1636038969321073 + ], + [ + -0.8314696123025453, + 0.0, + 0.5555702330196022, + 0.500013209717642, + 0.0, + 1.0, + 0.0, + 0.0, + -0.5555702330196022, + 0.0, + -0.8314696123025453, + 0.051677348927709255 + ], + [ + -0.9238795325112867, + 0.0, + 0.3826834323650899, + 0.34441508912858093, + 0.0, + 1.0, + 0.0, + 0.0, + -0.3826834323650899, + 0.0, + -0.9238795325112867, + -0.031491579260158 + ], + [ + -0.9807852804032304, + 0.0, + 0.1950903220161286, + 0.17558128981451576, + 0.0, + 1.0, + 0.0, + 0.0, + -0.1950903220161286, + 0.0, + -0.9807852804032304, + -0.08270675236290737 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_Round-RI_90.json b/examples/camera_poses/test_camera_Round-RI_90.json new file mode 100644 index 0000000000000000000000000000000000000000..a817879cc2b5650fbde1dbb20262086b7a7e0ad8 --- /dev/null +++ b/examples/camera_poses/test_camera_Round-RI_90.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 1.7000000000000002 + ], + [ + 0.9951847266721969, + 0.0, + 0.0980171403295606, + 0.08821542629660455, + 0.0, + 1.0, + 0.0, + 0.0, + -0.0980171403295606, + 0.0, + 0.9951847266721969, + 1.6956662540049772 + ], + [ + 0.9807852804032304, + 0.0, + 0.19509032201612825, + 0.17558128981451543, + 0.0, + 1.0, + 0.0, + 0.0, + -0.19509032201612825, + 0.0, + 0.9807852804032304, + 1.6827067523629076 + ], + [ + 0.9569403357322088, + 0.0, + 0.29028467725446233, + 0.2612562095290161, + 0.0, + 1.0, + 0.0, + 0.0, + -0.29028467725446233, + 0.0, + 0.9569403357322088, + 1.661246302158988 + ], + [ + 0.9238795325112867, + 0.0, + 0.3826834323650898, + 0.3444150891285808, + 0.0, + 1.0, + 0.0, + 0.0, + -0.3826834323650898, + 0.0, + 0.9238795325112867, + 1.631491579260158 + ], + [ + 0.881921264348355, + 0.0, + 0.47139673682599764, + 0.4242570631433979, + 0.0, + 1.0, + 0.0, + 0.0, + -0.47139673682599764, + 0.0, + 0.881921264348355, + 1.5937291379135194 + ], + [ + 0.8314696123025452, + 0.0, + 0.5555702330196022, + 0.500013209717642, + 0.0, + 1.0, + 0.0, + 0.0, + -0.5555702330196022, + 0.0, + 0.8314696123025452, + 1.5483226510722907 + ], + [ + 0.773010453362737, + 0.0, + 0.6343932841636455, + 0.5709539557472809, + 0.0, + 1.0, + 0.0, + 0.0, + -0.6343932841636455, + 0.0, + 0.773010453362737, + 1.4957094080264635 + ], + [ + 0.7071067811865476, + 0.0, + 0.7071067811865476, + 0.6363961030678928, + 0.0, + 1.0, + 0.0, + 0.0, + -0.7071067811865476, + 0.0, + 0.7071067811865476, + 1.436396103067893 + ], + [ + 0.6343932841636455, + 0.0, + 0.7730104533627369, + 0.6957094080264632, + 0.0, + 1.0, + 0.0, + 0.0, + -0.7730104533627369, + 0.0, + 0.6343932841636455, + 1.370953955747281 + ], + [ + 0.5555702330196023, + 0.0, + 0.8314696123025452, + 0.7483226510722907, + 0.0, + 1.0, + 0.0, + 0.0, + -0.8314696123025452, + 0.0, + 0.5555702330196023, + 1.3000132097176422 + ], + [ + 0.4713967368259978, + 0.0, + 0.8819212643483549, + 0.7937291379135195, + 0.0, + 1.0, + 0.0, + 0.0, + -0.8819212643483549, + 0.0, + 0.4713967368259978, + 1.2242570631433982 + ], + [ + 0.38268343236508984, + 0.0, + 0.9238795325112867, + 0.831491579260158, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9238795325112867, + 0.0, + 0.38268343236508984, + 1.144415089128581 + ], + [ + 0.29028467725446233, + 0.0, + 0.9569403357322089, + 0.861246302158988, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9569403357322089, + 0.0, + 0.29028467725446233, + 1.0612562095290161 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.8827067523629074, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.9755812898145155 + ], + [ + 0.09801714032956077, + 0.0, + 0.9951847266721968, + 0.8956662540049771, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9951847266721968, + 0.0, + 0.09801714032956077, + 0.8882154262966048 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_Round-ZoomIn.json b/examples/camera_poses/test_camera_Round-ZoomIn.json new file mode 100644 index 0000000000000000000000000000000000000000..85eccf2144a24125ba418c778ff7fb783a71f050 --- /dev/null +++ b/examples/camera_poses/test_camera_Round-ZoomIn.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.81 + ], + [ + 0.9807852804032304, + 0.0, + 0.19509032201612825, + 0.0019509032201612826, + 0.0, + 1.0, + 0.0, + 0.0, + -0.19509032201612825, + 0.0, + 0.9807852804032304, + 0.8098078528040323 + ], + [ + 0.9238795325112867, + 0.0, + 0.3826834323650898, + 0.003826834323650898, + 0.0, + 1.0, + 0.0, + 0.0, + -0.3826834323650898, + 0.0, + 0.9238795325112867, + 0.8092387953251129 + ], + [ + 0.8314696123025452, + 0.0, + 0.5555702330196022, + 0.005555702330196022, + 0.0, + 1.0, + 0.0, + 0.0, + -0.5555702330196022, + 0.0, + 0.8314696123025452, + 0.8083146961230255 + ], + [ + 0.7071067811865476, + 0.0, + 0.7071067811865476, + 0.007071067811865476, + 0.0, + 1.0, + 0.0, + 0.0, + -0.7071067811865476, + 0.0, + 0.7071067811865476, + 0.8070710678118656 + ], + [ + 0.5555702330196023, + 0.0, + 0.8314696123025452, + 0.008314696123025453, + 0.0, + 1.0, + 0.0, + 0.0, + -0.8314696123025452, + 0.0, + 0.5555702330196023, + 0.805555702330196 + ], + [ + 0.38268343236508984, + 0.0, + 0.9238795325112867, + 0.009238795325112868, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9238795325112867, + 0.0, + 0.38268343236508984, + 0.803826834323651 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.8019509032201614 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.8019509032201614 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.7019509032201614 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.6019509032201614 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.5019509032201613 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.4019509032201613 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.3019509032201613 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.2019509032201613 + ], + [ + 0.19509032201612833, + 0.0, + 0.9807852804032304, + 0.009807852804032305, + 0.0, + 1.0, + 0.0, + 0.0, + -0.9807852804032304, + 0.0, + 0.19509032201612833, + 0.10195090322016129 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_SPIN-ACW-60.json b/examples/camera_poses/test_camera_SPIN-ACW-60.json new file mode 100644 index 0000000000000000000000000000000000000000..647ab13af36b35c8a47ae2a6edc3154b3418b114 --- /dev/null +++ b/examples/camera_poses/test_camera_SPIN-ACW-60.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9978589232386035, + -0.06540312923014306, + 0.0, + 0.0, + 0.06540312923014306, + 0.9978589232386035, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9914448613738104, + -0.13052619222005157, + 0.0, + 0.0, + 0.13052619222005157, + 0.9914448613738104, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9807852804032304, + -0.19509032201612825, + 0.0, + 0.0, + 0.19509032201612825, + 0.9807852804032304, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9659258262890683, + -0.25881904510252074, + 0.0, + 0.0, + 0.25881904510252074, + 0.9659258262890683, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9469301294951057, + -0.32143946530316153, + 0.0, + 0.0, + 0.32143946530316153, + 0.9469301294951057, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9238795325112867, + -0.3826834323650898, + 0.0, + 0.0, + 0.3826834323650898, + 0.9238795325112867, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.8968727415326884, + -0.44228869021900125, + 0.0, + 0.0, + 0.44228869021900125, + 0.8968727415326884, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.8660254037844387, + -0.49999999999999994, + 0.0, + 0.0, + 0.49999999999999994, + 0.8660254037844387, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.8314696123025452, + -0.5555702330196022, + 0.0, + 0.0, + 0.5555702330196022, + 0.8314696123025452, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.7933533402912353, + -0.6087614290087205, + 0.0, + 0.0, + 0.6087614290087205, + 0.7933533402912353, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.7518398074789774, + -0.6593458151000688, + 0.0, + 0.0, + 0.6593458151000688, + 0.7518398074789774, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.7071067811865476, + -0.7071067811865476, + 0.0, + 0.0, + 0.7071067811865476, + 0.7071067811865476, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.659345815100069, + -0.7518398074789773, + 0.0, + 0.0, + 0.7518398074789773, + 0.659345815100069, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.6087614290087207, + -0.7933533402912352, + 0.0, + 0.0, + 0.7933533402912352, + 0.6087614290087207, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.5555702330196024, + -0.8314696123025451, + 0.0, + 0.0, + 0.8314696123025451, + 0.5555702330196024, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_SPIN-CW-60.json b/examples/camera_poses/test_camera_SPIN-CW-60.json new file mode 100644 index 0000000000000000000000000000000000000000..bab5e326a69c7ff99d562281bcf6873af0ff242b --- /dev/null +++ b/examples/camera_poses/test_camera_SPIN-CW-60.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9978589232386035, + 0.06540312923014306, + 0.0, + 0.0, + -0.06540312923014306, + 0.9978589232386035, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9914448613738104, + 0.13052619222005157, + 0.0, + 0.0, + -0.13052619222005157, + 0.9914448613738104, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9807852804032304, + 0.19509032201612825, + 0.0, + 0.0, + -0.19509032201612825, + 0.9807852804032304, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9659258262890683, + 0.25881904510252074, + 0.0, + 0.0, + -0.25881904510252074, + 0.9659258262890683, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9469301294951057, + 0.32143946530316153, + 0.0, + 0.0, + -0.32143946530316153, + 0.9469301294951057, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.9238795325112867, + 0.3826834323650898, + 0.0, + 0.0, + -0.3826834323650898, + 0.9238795325112867, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.8968727415326884, + 0.44228869021900125, + 0.0, + 0.0, + -0.44228869021900125, + 0.8968727415326884, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.8660254037844387, + 0.49999999999999994, + 0.0, + 0.0, + -0.49999999999999994, + 0.8660254037844387, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.8314696123025452, + 0.5555702330196022, + 0.0, + 0.0, + -0.5555702330196022, + 0.8314696123025452, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.7933533402912353, + 0.6087614290087205, + 0.0, + 0.0, + -0.6087614290087205, + 0.7933533402912353, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.7518398074789774, + 0.6593458151000688, + 0.0, + 0.0, + -0.6593458151000688, + 0.7518398074789774, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.7071067811865476, + 0.7071067811865476, + 0.0, + 0.0, + -0.7071067811865476, + 0.7071067811865476, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.659345815100069, + 0.7518398074789773, + 0.0, + 0.0, + -0.7518398074789773, + 0.659345815100069, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.6087614290087207, + 0.7933533402912352, + 0.0, + 0.0, + -0.7933533402912352, + 0.6087614290087207, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 0.5555702330196024, + 0.8314696123025451, + 0.0, + 0.0, + -0.8314696123025451, + 0.5555702330196024, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_U.json b/examples/camera_poses/test_camera_U.json new file mode 100644 index 0000000000000000000000000000000000000000..f63aabf1e10d52d9770727e75c659bf1ae2f3d16 --- /dev/null +++ b/examples/camera_poses/test_camera_U.json @@ -0,0 +1,226 @@ +[ + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.2, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.28750000000000003, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.37500000000000006, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.4625000000000001, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.55, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.6375000000000002, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.7250000000000001, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.8125000000000002, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.9000000000000001, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 0.9875000000000003, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 1.0750000000000002, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 1.1625000000000003, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 1.2500000000000002, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 1.3375000000000001, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 1.4250000000000003, + 0.0, + 0.0, + 1.0, + 0.0 + ], + [ + 1.0, + 0.0, + 0.0, + 0.0, + 0.0, + 1.0, + 0.0, + 1.5125000000000004, + 0.0, + 0.0, + 1.0, + 0.0 + ] +] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_b133a504fc90a2d1.json b/examples/camera_poses/test_camera_b133a504fc90a2d1.json new file mode 100644 index 0000000000000000000000000000000000000000..b9df09dde068f9316c0db51ead1ad6cdd3dc5628 --- /dev/null +++ b/examples/camera_poses/test_camera_b133a504fc90a2d1.json @@ -0,0 +1 @@ +[[0.9999999403953552, -3.2563843288535566e-10, 8.624932434919685e-10, -5.431840754965833e-09, -3.2563843288535566e-10, 1.0, -7.078895802870022e-10, -2.678919752696629e-09, 8.624932434919685e-10, -7.078895802870022e-10, 1.0, 7.303774474110014e-09], [0.9999998807907104, 0.00026603759033605456, -0.0003414043167140335, -0.030107486993074417, -0.0002659278397914022, 0.9999999403953552, 0.00032293255208060145, 0.011978899128735065, 0.0003414914826862514, -0.0003228419227525592, 0.9999999403953552, -0.07893969118595123], [0.9999994039535522, -0.0006866907933726907, -0.0008186722989194095, -0.050785429775714874, 0.0006877407431602478, 0.9999989867210388, 0.0012827541213482618, 0.023323407396674156, 0.0008177922572940588, -0.0012833180371671915, 0.9999988675117493, -0.1486724615097046], [0.999999463558197, -0.0005364732351154089, -0.0008647883078083396, -0.0673225075006485, 0.0005379383219406009, 0.9999984502792358, 0.0016958877677097917, 0.03169107437133789, 0.0008638793369755149, -0.0016963522648438811, 0.9999982118606567, -0.21477271616458893], [0.9999995231628418, -0.0005853328038938344, 0.000731398060452193, -0.08539554476737976, 0.0005843221442773938, 0.9999988675117493, 0.0013807439245283604, 0.037947509437799454, -0.0007322035962715745, -0.0013803173787891865, 0.9999987483024597, -0.2858566641807556], [0.9999935030937195, -0.0007727851625531912, 0.00352613371796906, -0.10591986775398254, 0.0007684561423957348, 0.9999989867210388, 0.0012290359009057283, 0.0426139310002327, -0.003527079476043582, -0.0012263187672942877, 0.999993085861206, -0.3645589053630829], [0.9999805688858032, -0.0012497249990701675, 0.006103655323386192, -0.12257411330938339, 0.0012419418198987842, 0.9999984502792358, 0.001278668874874711, 0.05143251642584801, -0.006105243694037199, -0.0012710647424682975, 0.9999805688858032, -0.44212815165519714], [0.9999697804450989, -0.001364423311315477, 0.0076533216051757336, -0.13916294276714325, 0.0013530527940019965, 0.9999980330467224, 0.0014907552395015955, 0.06076823174953461, -0.0076553407125175, -0.0014803558588027954, 0.9999696016311646, -0.5131306648254395], [0.9999656081199646, -0.001309191924519837, 0.00818517804145813, -0.1577269434928894, 0.0012902054004371166, 0.9999964833259583, 0.0023244714830070734, 0.07221967726945877, -0.008188189007341862, -0.0023138325195759535, 0.9999638199806213, -0.5706046223640442], [0.9999715089797974, -0.000632039678748697, 0.0075194560922682285, -0.1842648684978485, 0.0006101227481849492, 0.9999955892562866, 0.002916603581979871, 0.0780157595872879, -0.0075212628580629826, -0.0029119334649294615, 0.9999675154685974, -0.6207911372184753], [0.999971330165863, 9.749359742272645e-05, 0.007570990361273289, -0.2200128734111786, -0.00012177121971035376, 0.9999948740005493, 0.0032062893733382225, 0.07083319127559662, -0.007570638321340084, -0.003207120578736067, 0.9999662041664124, -0.671477735042572], [0.9999842047691345, 0.0011085873702540994, 0.005505275446921587, -0.265299528837204, -0.0011242963373661041, 0.9999953508377075, 0.002851187251508236, 0.06430258601903915, -0.00550208892673254, -0.002857332583516836, 0.9999808073043823, -0.7294802069664001], [0.9999891519546509, 0.0030521126464009285, -0.003507567336782813, -0.3109421133995056, -0.0030410285107791424, 0.9999904632568359, 0.003161099273711443, 0.06503432989120483, 0.003517185803502798, -0.0031503995414823294, 0.999988853931427, -0.7917969226837158], [0.9995473623275757, 0.00647346256300807, -0.029379382729530334, -0.3455933630466461, -0.006376372650265694, 0.9999739527702332, 0.0033971993252635, 0.06981948018074036, 0.02940060943365097, -0.003208327107131481, 0.9995625615119934, -0.8474093675613403], [0.9966378808021545, 0.011493867263197899, -0.08112204819917679, -0.3555213510990143, -0.011167873628437519, 0.9999276399612427, 0.004471173509955406, 0.0734858587384224, 0.08116757124662399, -0.00355018163099885, 0.9966941475868225, -0.9062771201133728], [0.9889613389968872, 0.01762101612985134, -0.14712226390838623, -0.33937495946884155, -0.01693679392337799, 0.999839186668396, 0.0059022014029324055, 0.0776127353310585, 0.14720259606838226, -0.0033452697098255157, 0.9891006946563721, -0.9548948407173157]] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_d9642c8efc01481d.json b/examples/camera_poses/test_camera_d9642c8efc01481d.json new file mode 100644 index 0000000000000000000000000000000000000000..50f031204bcc0bbde6758ad34f091e87b072ca45 --- /dev/null +++ b/examples/camera_poses/test_camera_d9642c8efc01481d.json @@ -0,0 +1 @@ +[[1.0, -1.4532828274127496e-09, 3.928266045782891e-10, -8.700192566379883e-09, -1.4532828274127496e-09, 1.0, -1.7456260048565042e-10, 9.261455491405002e-11, 3.928266045782891e-10, -1.7456260048565042e-10, 1.0, -1.5881864712241622e-08], [0.9850640296936035, -0.04082970321178436, 0.16727784276008606, 0.20439539849758148, 0.03502606600522995, 0.9986825585365295, 0.03750050067901611, -0.15185177326202393, -0.16858860850334167, -0.03108130767941475, 0.9851963520050049, 0.10121102631092072], [0.9168016910552979, -0.09482394903898239, 0.387921541929245, 0.44365406036376953, 0.06679922342300415, 0.9941277503967285, 0.08513444662094116, -0.26796984672546387, -0.39371633529663086, -0.0521385557949543, 0.9177521467208862, 0.06753730028867722], [0.7643709182739258, -0.15802493691444397, 0.6251122951507568, 0.6479341983795166, 0.09576083719730377, 0.9865723252296448, 0.13230620324611664, -0.33551496267318726, -0.6376261115074158, -0.04126973822712898, 0.7692397236824036, -0.07632352411746979], [0.5624101758003235, -0.2065712958574295, 0.8006392121315002, 0.8095709085464478, 0.10458854585886002, 0.9782856106758118, 0.17893710732460022, -0.3449772298336029, -0.8202170729637146, -0.016898371279239655, 0.5718027949333191, -0.267223596572876], [0.3860713243484497, -0.23765023052692413, 0.8913311958312988, 1.017128586769104, 0.10842984914779663, 0.9712379574775696, 0.21198998391628265, -0.34939509630203247, -0.9160742163658142, 0.014803661964833736, 0.4007355272769928, -0.5046621561050415], [0.25270596146583557, -0.2560441493988037, 0.9330493807792664, 1.2611535787582397, 0.09987718611955643, 0.9661006331443787, 0.23806333541870117, -0.36809343099594116, -0.9623742699623108, 0.03303031623363495, 0.26971235871315, -0.8164812922477722], [0.12896282970905304, -0.26742202043533325, 0.9549105167388916, 1.4084848165512085, 0.09037449955940247, 0.9621138572692871, 0.25723403692245483, -0.3597045838832855, -0.9875227212905884, 0.05312592536211014, 0.14824506640434265, -1.1886308193206787], [-0.037818778306245804, -0.2752484083175659, 0.9606289863586426, 1.3755683898925781, 0.08146493136882782, 0.957267701625824, 0.277492493391037, -0.353816956281662, -0.9959584474563599, 0.08875200152397156, -0.013779602944850922, -1.6891316175460815], [-0.1970304548740387, -0.2752428948879242, 0.9409677982330322, 1.1786020994186401, 0.07280438393354416, 0.9530242681503296, 0.2940141260623932, -0.36850038170814514, -0.9776904582977295, 0.12643630802631378, -0.16773590445518494, -2.261430025100708], [-0.3677733540534973, -0.26668044924736023, 0.8908559679985046, 0.8089978098869324, 0.06243090331554413, 0.9487544894218445, 0.30978602170944214, -0.37012383341789246, -0.9278174042701721, 0.16954797506332397, -0.33227747678756714, -2.8139760494232178], [-0.5150132775306702, -0.2527574300765991, 0.8190696835517883, 0.4462648332118988, 0.050335872918367386, 0.9449706673622131, 0.32325950264930725, -0.381427526473999, -0.8557030558586121, 0.20771150290966034, -0.473949670791626, -3.2886314392089844], [-0.6352092623710632, -0.23657281696796417, 0.7352160215377808, 0.0826139897108078, 0.036419764161109924, 0.9416990280151367, 0.33447936177253723, -0.3719184398651123, -0.771480917930603, 0.23924078047275543, -0.5895600318908691, -3.677316904067993], [-0.7218965888023376, -0.21918001770973206, 0.6563730239868164, -0.2607730031013489, 0.024064550176262856, 0.9399895071983337, 0.340353786945343, -0.359468549489975, -0.6915825009346008, 0.26149556040763855, -0.6733006834983826, -3.9900803565979004], [-0.7831082344055176, -0.2019210308790207, 0.5881916880607605, -0.609000563621521, 0.015634668990969658, 0.9391284584999084, 0.3432103097438812, -0.35386255383491516, -0.621688961982727, 0.2779669761657715, -0.732282280921936, -4.316582202911377], [-0.8348898887634277, -0.1848600059747696, 0.5184455513954163, -0.9689992070198059, 0.007868430577218533, 0.93780916929245, 0.3470619320869446, -0.3402447998523712, -0.5503608584403992, 0.2938378155231476, -0.7815127968788147, -4.561704635620117]] \ No newline at end of file diff --git a/examples/camera_poses/test_camera_d971457c81bca597.json b/examples/camera_poses/test_camera_d971457c81bca597.json new file mode 100644 index 0000000000000000000000000000000000000000..3bd9bf5dd6993fd3b7fe6d678595060378a3c997 --- /dev/null +++ b/examples/camera_poses/test_camera_d971457c81bca597.json @@ -0,0 +1 @@ +[[1.0, -1.9455232563858615e-11, -8.611562019034125e-10, -1.0473515388298438e-09, -1.9455232563858615e-11, 1.0, 8.674055917978762e-10, 1.8375034827045056e-09, -8.611562019034125e-10, 8.674055917978762e-10, 1.0, 8.45981773522908e-09], [0.9973401427268982, -0.0032562706619501114, 0.07281485944986343, -0.07197967171669006, 0.00329946493729949, 0.9999944567680359, -0.00047293686657212675, -0.0565447174012661, -0.07281292229890823, 0.0007119301008060575, 0.9973453879356384, -0.3188611567020416], [0.9906007051467896, -0.0066957552917301655, 0.13662146031856537, -0.15415722131729126, 0.006810691673308611, 0.9999766945838928, -0.00037385127507150173, -0.10014614462852478, -0.1366157978773117, 0.0013008243404328823, 0.9906232953071594, -0.6077051758766174], [0.9827210903167725, -0.009423106908798218, 0.18485280871391296, -0.23053960502147675, 0.009786692447960377, 0.9999515414237976, -0.0010545575059950352, -0.11272216588258743, -0.18483391404151917, 0.002845433307811618, 0.9827656745910645, -0.8897562623023987], [0.9756309986114502, -0.011318936944007874, 0.2191258817911148, -0.2929331362247467, 0.01199379749596119, 0.9999265670776367, -0.0017497432418167591, -0.10348402708768845, -0.21908999979496002, 0.004335256293416023, 0.9756950736045837, -1.1368639469146729], [0.9685831069946289, -0.012191502377390862, 0.24839124083518982, -0.379209041595459, 0.013217172585427761, 0.9999096393585205, -0.0024619612377136946, -0.12078691273927689, -0.24833877384662628, 0.005667644087225199, 0.9686566591262817, -1.3666095733642578], [0.9615083932876587, -0.012519675306975842, 0.2744903266429901, -0.5242342352867126, 0.013806014321744442, 0.9999009370803833, -0.0027547888457775116, -0.13015854358673096, -0.2744286358356476, 0.006438371259719133, 0.9615859389305115, -1.5762972831726074], [0.9528889656066895, -0.014083069749176502, 0.3029923439025879, -0.6702165007591248, 0.015383570455014706, 0.9998798370361328, -0.0019058430334553123, -0.14860114455223083, -0.3029291331768036, 0.00647716224193573, 0.9529911279678345, -1.7173497676849365], [0.9457509517669678, -0.015159872360527515, 0.3245387375354767, -0.8001917600631714, 0.016729505732655525, 0.9998579621315002, -0.0020466779824346304, -0.15521110594272614, -0.32446160912513733, 0.007365020923316479, 0.9458702206611633, -1.8571592569351196], [0.9417706727981567, -0.015540024265646935, 0.33589670062065125, -0.9244012236595154, 0.017394419759511948, 0.9998455047607422, -0.0025124624371528625, -0.1416105479001999, -0.3358057737350464, 0.008208892308175564, 0.9418954849243164, -1.9969842433929443], [0.9381415247917175, -0.016401933506131172, 0.34586337208747864, -1.0128529071807861, 0.01851990446448326, 0.9998245239257812, -0.0028197101783007383, -0.11503250896930695, -0.3457564413547516, 0.009050644934177399, 0.9382806420326233, -2.080850124359131], [0.9357938766479492, -0.016873901709914207, 0.3521435558795929, -1.1008079051971436, 0.01865064539015293, 0.9998247027397156, -0.0016533475136384368, -0.048780668526887894, -0.3520539104938507, 0.008114897646009922, 0.9359445571899414, -2.088500499725342], [0.9334627985954285, -0.016492612659931183, 0.3582947850227356, -1.2417148351669312, 0.017931735143065453, 0.9998390078544617, -0.0006939812446944416, 0.01625901833176613, -0.35822567343711853, 0.007072653621435165, 0.9336082339286804, -2.0250792503356934], [0.9281281232833862, -0.016213543713092804, 0.3719078302383423, -1.4365860223770142, 0.01746317930519581, 0.9998475313186646, 8.084020009846427e-06, 0.060201361775398254, -0.37185123562812805, 0.006487190257757902, 0.9282696843147278, -1.8963005542755127], [0.9220678806304932, -0.017285069450736046, 0.3866421580314636, -1.604330062866211, 0.01880805380642414, 0.9998230934143066, -0.00015593231364618987, 0.0799601748585701, -0.38657107949256897, 0.007415766827762127, 0.9222298264503479, -1.7388361692428589], [0.9185494184494019, -0.018356993794441223, 0.3948797583580017, -1.7530856132507324, 0.019847355782985687, 0.9998030066490173, 0.0003104731731582433, 0.10000917315483093, -0.3948076665401459, 0.007552135270088911, 0.9187328219413757, -1.5337872505187988]] \ No newline at end of file diff --git a/examples/trajectories/curve_1.txt b/examples/trajectories/curve_1.txt new file mode 100644 index 0000000000000000000000000000000000000000..87e0ef6929bf9dbb1cba7f2bd12ce8a8ab5d592e --- /dev/null +++ b/examples/trajectories/curve_1.txt @@ -0,0 +1,59 @@ +88, 173 +89, 173 +91, 173 +92, 173 +93, 173 +93, 174 +95, 175 +97, 176 +98, 176 +100, 178 +103, 180 +104, 181 +105, 181 +106, 182 +107, 182 +108, 182 +110, 184 +111, 185 +113, 186 +115, 186 +116, 187 +116, 188 +119, 189 +122, 190 +125, 191 +128, 192 +129, 192 +130, 192 +133, 193 +134, 193 +140, 194 +146, 194 +150, 194 +151, 194 +153, 194 +155, 194 +156, 194 +157, 194 +158, 194 +159, 194 +162, 193 +165, 192 +168, 191 +169, 190 +170, 190 +172, 189 +173, 188 +173, 187 +174, 187 +175, 186 +178, 185 +179, 183 +180, 183 +181, 183 +183, 181 +185, 180 +185, 179 +185, 179 +185, 179 diff --git a/examples/trajectories/curve_2.txt b/examples/trajectories/curve_2.txt new file mode 100644 index 0000000000000000000000000000000000000000..f1b896e725a273156d0fe572337f4c7bffcf1c41 --- /dev/null +++ b/examples/trajectories/curve_2.txt @@ -0,0 +1,80 @@ +206, 146 +206, 147 +205, 149 +204, 151 +203, 151 +202, 154 +201, 156 +200, 159 +199, 161 +198, 163 +197, 164 +197, 165 +197, 166 +195, 168 +193, 171 +191, 173 +189, 176 +188, 177 +187, 178 +187, 179 +186, 180 +185, 181 +184, 181 +184, 182 +182, 184 +180, 186 +179, 187 +177, 189 +175, 191 +174, 191 +173, 192 +172, 193 +171, 193 +169, 195 +167, 197 +164, 198 +161, 200 +160, 201 +158, 202 +156, 204 +155, 204 +154, 204 +153, 205 +151, 206 +150, 206 +149, 206 +147, 207 +146, 207 +143, 209 +139, 210 +138, 210 +135, 211 +135, 212 +134, 212 +133, 212 +132, 212 +131, 212 +130, 212 +129, 212 +128, 212 +127, 212 +126, 213 +124, 213 +122, 213 +120, 213 +120, 214 +119, 214 +118, 214 +116, 214 +115, 214 +114, 214 +113, 214 +112, 214 +111, 214 +110, 214 +110, 215 +109, 215 +108, 215 +108, 215 +108, 215 diff --git a/examples/trajectories/curve_3.txt b/examples/trajectories/curve_3.txt new file mode 100644 index 0000000000000000000000000000000000000000..0456a0fbe712a29bb3e4c7edc46b678c64221f11 --- /dev/null +++ b/examples/trajectories/curve_3.txt @@ -0,0 +1,150 @@ +73, 149 +74, 149 +76, 149 +82, 149 +83, 149 +90, 149 +98, 149 +102, 149 +104, 149 +105, 149 +106, 150 +108, 152 +111, 154 +114, 157 +116, 158 +117, 160 +118, 161 +119, 164 +121, 166 +122, 168 +122, 169 +123, 171 +123, 172 +124, 174 +125, 177 +126, 179 +126, 181 +126, 183 +126, 184 +126, 185 +127, 186 +127, 187 +127, 188 +127, 189 +127, 190 +127, 191 +128, 192 +128, 193 +129, 194 +130, 196 +130, 197 +131, 198 +131, 199 +134, 201 +135, 202 +137, 203 +138, 204 +139, 204 +139, 205 +140, 206 +142, 206 +142, 207 +147, 209 +151, 211 +153, 212 +154, 212 +156, 212 +158, 213 +159, 214 +161, 214 +162, 214 +163, 214 +164, 214 +165, 214 +167, 214 +168, 214 +169, 214 +170, 214 +171, 214 +172, 213 +173, 213 +175, 212 +176, 212 +177, 212 +178, 211 +179, 211 +179, 208 +182, 205 +184, 203 +184, 202 +186, 200 +187, 199 +187, 198 +188, 196 +190, 194 +192, 191 +193, 189 +193, 188 +194, 188 +194, 187 +195, 184 +196, 183 +197, 182 +197, 180 +198, 180 +198, 179 +199, 178 +199, 177 +200, 176 +200, 175 +201, 174 +202, 173 +203, 171 +204, 170 +204, 169 +205, 169 +205, 167 +207, 165 +208, 163 +209, 163 +209, 162 +210, 162 +211, 162 +212, 162 +213, 162 +215, 162 +216, 162 +217, 163 +219, 164 +220, 164 +221, 165 +223, 166 +224, 166 +224, 167 +225, 167 +225, 168 +226, 169 +227, 171 +228, 173 +228, 174 +229, 175 +229, 176 +229, 177 +229, 180 +229, 181 +229, 182 +229, 183 +229, 184 +229, 185 +230, 186 +230, 187 +230, 187 +230, 189 +224, 192 +223, 193 +220, 194 +219, 195 +218, 195 +218, 195 +218, 195 diff --git a/examples/trajectories/curve_4.txt b/examples/trajectories/curve_4.txt new file mode 100644 index 0000000000000000000000000000000000000000..e9404339b1006c3b63fbe11a797ab76e39a5b996 --- /dev/null +++ b/examples/trajectories/curve_4.txt @@ -0,0 +1,137 @@ +176, 109 +177, 110 +177, 111 +177, 112 +177, 114 +178, 118 +179, 122 +180, 125 +180, 126 +180, 127 +180, 128 +180, 129 +180, 130 +180, 131 +180, 133 +180, 134 +179, 136 +177, 138 +177, 139 +176, 140 +176, 141 +174, 143 +172, 145 +170, 147 +170, 148 +167, 150 +163, 154 +159, 158 +158, 158 +155, 160 +154, 161 +152, 163 +151, 164 +149, 165 +147, 166 +145, 167 +143, 169 +140, 171 +137, 172 +135, 174 +133, 175 +131, 176 +130, 176 +128, 177 +127, 177 +125, 178 +123, 179 +121, 179 +119, 179 +116, 179 +114, 180 +112, 180 +111, 180 +110, 180 +108, 180 +104, 181 +99, 181 +95, 181 +94, 181 +93, 179 +92, 179 +92, 176 +91, 173 +91, 171 +91, 170 +91, 168 +91, 165 +91, 164 +91, 163 +91, 162 +91, 161 +93, 159 +94, 159 +99, 156 +100, 156 +105, 153 +106, 152 +110, 149 +112, 148 +113, 148 +114, 147 +115, 146 +117, 145 +119, 144 +120, 144 +121, 143 +122, 142 +123, 142 +124, 142 +125, 141 +126, 141 +127, 140 +128, 140 +128, 139 +129, 138 +128, 138 +126, 137 +125, 136 +123, 136 +122, 136 +121, 135 +120, 135 +119, 135 +118, 135 +117, 135 +116, 135 +115, 135 +112, 134 +111, 134 +109, 134 +107, 133 +104, 133 +103, 133 +102, 133 +100, 133 +99, 133 +96, 133 +94, 133 +93, 133 +92, 133 +91, 133 +89, 133 +88, 133 +87, 133 +86, 133 +85, 133 +84, 133 +82, 133 +81, 133 +80, 133 +79, 133 +78, 133 +77, 133 +76, 133 +76, 133 +76, 133 +76, 133 diff --git a/examples/trajectories/horizon_2.txt b/examples/trajectories/horizon_2.txt new file mode 100644 index 0000000000000000000000000000000000000000..6b2cc3c3783050131fbfe44450cf615fe6f8ab40 --- /dev/null +++ b/examples/trajectories/horizon_2.txt @@ -0,0 +1,48 @@ +121, 124 +123, 124 +124, 124 +125, 124 +127, 124 +130, 124 +131, 124 +135, 124 +140, 124 +143, 124 +144, 124 +145, 124 +149, 124 +152, 124 +154, 124 +155, 124 +156, 124 +159, 125 +160, 125 +163, 125 +166, 126 +168, 126 +169, 126 +170, 126 +173, 126 +176, 126 +177, 126 +178, 126 +179, 126 +180, 126 +181, 126 +183, 126 +184, 126 +185, 126 +186, 126 +187, 126 +188, 126 +189, 126 +190, 126 +191, 126 +192, 126 +193, 126 +194, 126 +195, 126 +196, 126 +196, 126 +196, 126 +196, 126 diff --git a/examples/trajectories/shake_1.txt b/examples/trajectories/shake_1.txt new file mode 100644 index 0000000000000000000000000000000000000000..c52500d294a4b29e7e19853395aca59124282505 --- /dev/null +++ b/examples/trajectories/shake_1.txt @@ -0,0 +1,205 @@ +103, 89 +104, 89 +106, 89 +107, 89 +108, 89 +109, 89 +110, 89 +111, 89 +112, 89 +113, 89 +114, 89 +115, 89 +116, 89 +117, 89 +118, 89 +119, 89 +120, 89 +122, 89 +123, 89 +124, 89 +125, 89 +126, 89 +127, 88 +128, 88 +129, 88 +130, 88 +131, 88 +133, 87 +136, 86 +137, 86 +138, 86 +139, 86 +140, 86 +141, 86 +142, 86 +143, 86 +144, 86 +145, 86 +146, 87 +147, 87 +148, 87 +149, 87 +148, 87 +146, 87 +145, 88 +144, 88 +142, 89 +141, 89 +140, 90 +140, 91 +138, 91 +137, 92 +136, 92 +136, 93 +135, 93 +134, 93 +133, 93 +132, 93 +131, 93 +130, 93 +129, 93 +128, 93 +127, 92 +125, 92 +124, 92 +123, 92 +122, 92 +121, 92 +120, 92 +119, 92 +118, 92 +117, 92 +116, 92 +115, 92 +113, 92 +112, 92 +111, 92 +110, 92 +109, 92 +108, 92 +108, 91 +108, 90 +109, 90 +110, 90 +111, 89 +112, 89 +113, 89 +114, 89 +115, 89 +115, 88 +116, 88 +117, 88 +118, 88 +118, 87 +119, 87 +120, 87 +121, 87 +122, 86 +123, 86 +124, 86 +125, 86 +126, 85 +127, 85 +128, 85 +129, 85 +130, 85 +131, 85 +132, 85 +133, 85 +134, 85 +135, 85 +136, 85 +137, 85 +138, 85 +139, 85 +140, 85 +141, 85 +142, 85 +143, 85 +143, 84 +144, 84 +145, 84 +146, 84 +147, 84 +148, 84 +149, 84 +148, 84 +147, 84 +145, 84 +144, 84 +143, 84 +142, 84 +141, 84 +140, 85 +139, 85 +138, 85 +137, 86 +136, 86 +136, 87 +135, 87 +134, 87 +133, 87 +132, 88 +131, 88 +130, 88 +129, 88 +129, 89 +128, 89 +127, 89 +126, 89 +125, 89 +124, 90 +123, 90 +122, 90 +121, 90 +120, 91 +119, 91 +118, 91 +117, 91 +116, 91 +115, 91 +114, 91 +113, 91 +112, 91 +111, 91 +110, 91 +109, 91 +109, 90 +108, 90 +110, 90 +111, 90 +113, 90 +114, 90 +115, 90 +116, 90 +118, 90 +120, 90 +121, 90 +122, 90 +123, 90 +124, 90 +126, 90 +127, 90 +128, 90 +129, 90 +130, 90 +131, 90 +132, 90 +133, 90 +134, 90 +135, 90 +136, 90 +137, 90 +138, 90 +139, 90 +140, 90 +141, 89 +142, 89 +143, 89 +144, 89 +145, 89 +146, 89 +147, 89 +147, 89 +147, 89 diff --git a/examples/trajectories/shake_2.txt b/examples/trajectories/shake_2.txt new file mode 100644 index 0000000000000000000000000000000000000000..d8a5bcc9980468d3ac5ae052761cb2dc3d2c7d9d --- /dev/null +++ b/examples/trajectories/shake_2.txt @@ -0,0 +1,156 @@ +125, 86 +127, 86 +129, 86 +130, 86 +131, 86 +133, 86 +134, 86 +135, 86 +137, 86 +138, 86 +139, 86 +140, 86 +141, 86 +142, 86 +141, 86 +139, 86 +139, 87 +138, 87 +137, 87 +136, 87 +135, 87 +134, 87 +134, 88 +133, 88 +132, 88 +131, 88 +130, 88 +129, 88 +128, 88 +127, 88 +128, 88 +129, 88 +130, 88 +131, 88 +132, 88 +133, 88 +134, 88 +135, 88 +136, 88 +137, 88 +138, 88 +139, 88 +140, 88 +141, 88 +140, 88 +139, 88 +138, 88 +136, 88 +134, 88 +133, 88 +131, 88 +130, 88 +129, 88 +128, 88 +127, 88 +128, 88 +129, 88 +130, 88 +132, 87 +133, 87 +134, 86 +136, 86 +137, 86 +138, 85 +139, 85 +140, 85 +140, 84 +141, 84 +139, 85 +138, 85 +137, 86 +136, 86 +135, 86 +134, 86 +133, 86 +132, 86 +131, 86 +130, 86 +129, 86 +128, 86 +127, 86 +126, 86 +127, 86 +129, 86 +130, 86 +132, 86 +133, 86 +135, 86 +137, 86 +138, 86 +139, 86 +140, 86 +141, 87 +142, 87 +143, 87 +144, 88 +143, 88 +142, 88 +141, 88 +139, 88 +138, 88 +137, 88 +136, 88 +135, 88 +134, 88 +133, 88 +131, 88 +130, 88 +129, 88 +128, 88 +127, 88 +128, 88 +129, 88 +130, 88 +132, 88 +133, 88 +134, 88 +136, 88 +137, 88 +138, 88 +139, 88 +140, 88 +141, 88 +142, 88 +143, 88 +144, 88 +143, 88 +142, 88 +141, 88 +140, 88 +139, 88 +138, 88 +137, 88 +135, 89 +135, 90 +133, 90 +131, 91 +130, 91 +129, 91 +128, 91 +127, 91 +128, 91 +129, 91 +130, 91 +132, 91 +133, 90 +134, 89 +135, 89 +136, 89 +137, 89 +138, 89 +138, 88 +139, 88 +140, 88 +140, 88 +140, 88 diff --git a/examples/trajectories/shaking_10.txt b/examples/trajectories/shaking_10.txt new file mode 100644 index 0000000000000000000000000000000000000000..939933d7a4d02594c502000c559b77526f961d0d --- /dev/null +++ b/examples/trajectories/shaking_10.txt @@ -0,0 +1,373 @@ +117, 102 +114, 102 +109, 102 +106, 102 +105, 102 +102, 102 +99, 102 +97, 102 +96, 102 +95, 102 +93, 102 +89, 102 +85, 103 +82, 103 +81, 103 +80, 103 +79, 103 +78, 103 +76, 103 +74, 104 +73, 104 +72, 104 +71, 104 +70, 105 +69, 105 +68, 105 +67, 105 +66, 106 +64, 107 +63, 108 +62, 108 +61, 108 +61, 109 +60, 109 +59, 109 +58, 109 +57, 110 +56, 110 +55, 111 +54, 111 +53, 111 +52, 111 +52, 112 +51, 112 +50, 112 +50, 113 +49, 113 +48, 113 +46, 114 +46, 115 +45, 115 +45, 116 +44, 116 +43, 117 +42, 117 +41, 117 +41, 118 +40, 118 +41, 118 +41, 119 +42, 119 +43, 119 +44, 119 +46, 119 +47, 119 +48, 119 +49, 119 +50, 119 +51, 119 +52, 119 +53, 119 +54, 119 +55, 119 +56, 118 +58, 118 +59, 118 +61, 118 +63, 118 +64, 117 +67, 117 +70, 117 +71, 117 +73, 117 +75, 116 +76, 116 +77, 116 +80, 116 +82, 116 +83, 116 +84, 116 +85, 116 +88, 116 +91, 116 +94, 116 +97, 116 +98, 116 +100, 116 +101, 117 +102, 117 +104, 117 +105, 117 +106, 117 +107, 117 +108, 117 +109, 117 +110, 117 +111, 117 +115, 117 +119, 117 +123, 117 +124, 117 +128, 117 +129, 117 +132, 117 +134, 117 +135, 117 +136, 117 +138, 117 +139, 117 +140, 117 +141, 117 +142, 116 +145, 116 +146, 116 +148, 116 +149, 116 +151, 115 +152, 115 +153, 115 +154, 115 +155, 114 +156, 114 +157, 114 +158, 114 +159, 114 +162, 114 +163, 113 +164, 113 +165, 113 +166, 113 +167, 113 +168, 113 +169, 113 +170, 113 +171, 113 +172, 113 +173, 113 +174, 113 +175, 113 +178, 113 +181, 113 +182, 113 +183, 113 +184, 113 +185, 113 +187, 113 +188, 113 +189, 113 +191, 113 +192, 113 +193, 113 +194, 113 +195, 113 +196, 113 +197, 113 +198, 113 +199, 113 +200, 113 +201, 113 +202, 113 +203, 113 +202, 113 +201, 113 +200, 113 +198, 113 +197, 113 +196, 113 +195, 112 +194, 112 +193, 112 +192, 112 +191, 111 +190, 111 +189, 111 +188, 110 +187, 110 +186, 110 +185, 110 +184, 110 +183, 110 +182, 110 +181, 110 +180, 110 +179, 110 +178, 110 +177, 110 +175, 110 +173, 110 +172, 110 +171, 110 +170, 110 +168, 110 +167, 110 +165, 110 +164, 110 +163, 110 +161, 111 +159, 111 +155, 111 +153, 111 +151, 111 +151, 112 +150, 112 +149, 112 +148, 112 +147, 112 +145, 112 +143, 113 +142, 113 +140, 113 +139, 113 +138, 113 +136, 113 +135, 113 +134, 113 +133, 114 +131, 114 +130, 114 +128, 115 +127, 115 +126, 115 +125, 115 +124, 115 +122, 115 +121, 115 +120, 115 +118, 116 +115, 116 +113, 116 +111, 116 +109, 117 +106, 117 +103, 117 +102, 117 +100, 117 +98, 117 +97, 117 +95, 117 +94, 117 +93, 117 +92, 117 +91, 117 +90, 117 +89, 117 +88, 117 +87, 117 +86, 117 +85, 117 +84, 117 +83, 117 +84, 117 +85, 117 +87, 117 +88, 117 +89, 117 +90, 117 +92, 117 +93, 117 +95, 117 +97, 117 +99, 117 +101, 117 +103, 117 +104, 117 +105, 117 +106, 117 +107, 117 +108, 117 +109, 117 +110, 117 +112, 117 +113, 117 +114, 117 +116, 117 +117, 117 +118, 117 +119, 117 +120, 117 +121, 117 +123, 117 +124, 117 +125, 117 +126, 117 +127, 117 +129, 117 +130, 117 +131, 117 +133, 117 +134, 117 +135, 117 +136, 117 +137, 117 +138, 117 +139, 117 +140, 117 +141, 117 +142, 117 +143, 117 +145, 117 +146, 117 +147, 117 +148, 117 +149, 117 +150, 117 +149, 117 +148, 117 +147, 117 +146, 117 +144, 117 +143, 118 +142, 118 +141, 118 +140, 118 +139, 118 +138, 118 +136, 118 +135, 118 +132, 119 +131, 119 +130, 119 +129, 119 +127, 119 +126, 119 +124, 119 +123, 119 +122, 119 +121, 119 +119, 119 +118, 119 +117, 119 +115, 119 +114, 119 +113, 119 +112, 119 +111, 119 +110, 119 +109, 119 +108, 119 +107, 119 +106, 119 +107, 119 +108, 119 +109, 119 +110, 119 +112, 119 +113, 119 +114, 119 +115, 119 +116, 119 +117, 119 +118, 119 +119, 119 +120, 119 +121, 119 +122, 119 +123, 119 +124, 119 +125, 119 +126, 119 +127, 119 +127, 119 +127, 119 +127, 119 diff --git a/gradio_utils/camera_utils.py b/gradio_utils/camera_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..2162e43b46458a62766335927e8c66b8451fb5de --- /dev/null +++ b/gradio_utils/camera_utils.py @@ -0,0 +1,146 @@ +import copy +# import plotly.express as px +# import plotly.graph_objects as go +import json + +import numpy as np + +CAMERA_MOTION_MODE = ["Basic Camera Poses", "Provided Complex Camera Poses", "Customize Camera Poses"] + +CAMERA = { + # T + "base_T_norm": 1.5, + "base_angle": np.pi/3, + + "Pan Up": { "angle":[0., 0., 0.], "T":[0., 1., 0.]}, + "Pan Down": { "angle":[0., 0., 0.], "T":[0.,-1.,0.]}, + "Pan Left": { "angle":[0., 0., 0.], "T":[1.,0.,0.]}, + "Pan Right": { "angle":[0., 0., 0.], "T": [-1.,0.,0.]}, + "Zoom In": { "angle":[0., 0., 0.], "T": [0.,0.,-2.]}, + "Zoom Out": { "angle":[0., 0., 0.], "T": [0.,0.,2.]}, + "ACW": { "angle": [0., 0., 1.], "T":[0., 0., 0.]}, + "CW": { "angle": [0., 0., -1.], "T":[0., 0., 0.]}, +} + +COMPLEX_CAMERA = { + "Pose_1": "examples/camera_poses/test_camera_1424acd0007d40b5.json", + "Pose_2": "examples/camera_poses/test_camera_d971457c81bca597.json", + "Pose_3": "examples/camera_poses/test_camera_Round-ZoomIn.json", + "Pose_4": "examples/camera_poses/test_camera_Round-RI_90.json", + "Pose_5": "examples/camera_poses/test_camera_Round-RI_120.json", + "Pose_6": "examples/camera_poses/test_camera_018f7907401f2fef.json", + "Pose_7": "examples/camera_poses/test_camera_088b93f15ca8745d.json", + "Pose_8": "examples/camera_poses/test_camera_b133a504fc90a2d1.json", +} + + + +def compute_R_form_rad_angle(angles): + theta_x, theta_y, theta_z = angles + Rx = np.array([[1, 0, 0], + [0, np.cos(theta_x), -np.sin(theta_x)], + [0, np.sin(theta_x), np.cos(theta_x)]]) + + Ry = np.array([[np.cos(theta_y), 0, np.sin(theta_y)], + [0, 1, 0], + [-np.sin(theta_y), 0, np.cos(theta_y)]]) + + Rz = np.array([[np.cos(theta_z), -np.sin(theta_z), 0], + [np.sin(theta_z), np.cos(theta_z), 0], + [0, 0, 1]]) + + # 计算相机外参的旋转矩阵 + R = np.dot(Rz, np.dot(Ry, Rx)) + return R + +def get_camera_motion(angle, T, speed, n=16): + RT = [] + for i in range(n): + _angle = (i/n)*speed*(CAMERA["base_angle"])*angle + R = compute_R_form_rad_angle(_angle) + # _T = (i/n)*speed*(T.reshape(3,1)) + _T=(i/n)*speed*(CAMERA["base_T_norm"])*(T.reshape(3,1)) + _RT = np.concatenate([R,_T], axis=1) + RT.append(_RT) + RT = np.stack(RT) + return RT + +def create_relative(RT_list, K_1=4.7, dataset="syn"): + RT = copy.deepcopy(RT_list[0]) + R_inv = RT[:,:3].T + T = RT[:,-1] + + temp = [] + for _RT in RT_list: + _RT[:,:3] = np.dot(_RT[:,:3], R_inv) + _RT[:,-1] = _RT[:,-1] - np.dot(_RT[:,:3], T) + temp.append(_RT) + RT_list = temp + + return RT_list + +def combine_camera_motion(RT_0, RT_1): + RT = copy.deepcopy(RT_0[-1]) + R = RT[:,:3] + R_inv = RT[:,:3].T + T = RT[:,-1] + + temp = [] + for _RT in RT_1: + _RT[:,:3] = np.dot(_RT[:,:3], R) + _RT[:,-1] = _RT[:,-1] + np.dot(np.dot(_RT[:,:3], R_inv), T) + temp.append(_RT) + + RT_1 = np.stack(temp) + + return np.concatenate([RT_0, RT_1], axis=0) + +def process_camera(camera_dict): + # "First A then B", "Both A and B", "Custom" + if camera_dict['complex'] is not None: + with open(COMPLEX_CAMERA[camera_dict['complex']]) as f: + RT = json.load(f) # [16, 12] + RT = np.array(RT).reshape(-1, 3, 4) + print(RT.shape) + return RT + + + motion_list = camera_dict['motion'] + mode = camera_dict['mode'] + speed = camera_dict['speed'] + print(len(motion_list)) + if len(motion_list) == 0: + angle = np.array([0,0,0]) + T = np.array([0,0,0]) + RT = get_camera_motion(angle, T, speed, 16) + + + elif len(motion_list) == 1: + angle = np.array(CAMERA[motion_list[0]]["angle"]) + T = np.array(CAMERA[motion_list[0]]["T"]) + print(angle, T) + RT = get_camera_motion(angle, T, speed, 16) + + + + elif len(motion_list) == 2: + if mode == "Customized Mode 1: First A then B": + angle = np.array(CAMERA[motion_list[0]]["angle"]) + T = np.array(CAMERA[motion_list[0]]["T"]) + RT_0 = get_camera_motion(angle, T, speed, 8) + + angle = np.array(CAMERA[motion_list[1]]["angle"]) + T = np.array(CAMERA[motion_list[1]]["T"]) + RT_1 = get_camera_motion(angle, T, speed, 8) + + RT = combine_camera_motion(RT_0, RT_1) + + elif mode == "Customized Mode 2: Both A and B": + angle = np.array(CAMERA[motion_list[0]]["angle"]) + np.array(CAMERA[motion_list[1]]["angle"]) + T = np.array(CAMERA[motion_list[0]]["T"]) + np.array(CAMERA[motion_list[1]]["T"]) + RT = get_camera_motion(angle, T, speed, 16) + + + # return RT.reshape(-1, 12) + return RT + diff --git a/gradio_utils/flow_utils.py b/gradio_utils/flow_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..3a86b38a3f555650fe6dd0f6d504bd6d5bbbf933 --- /dev/null +++ b/gradio_utils/flow_utils.py @@ -0,0 +1,69 @@ +import numpy as np + + +def sigma_matrix2(sig_x, sig_y, theta): + """Calculate the rotated sigma matrix (two dimensional matrix). + Args: + sig_x (float): + sig_y (float): + theta (float): Radian measurement. + Returns: + ndarray: Rotated sigma matrix. + """ + d_matrix = np.array([[sig_x**2, 0], [0, sig_y**2]]) + u_matrix = np.array([[np.cos(theta), -np.sin(theta)], [np.sin(theta), np.cos(theta)]]) + return np.dot(u_matrix, np.dot(d_matrix, u_matrix.T)) + + +def mesh_grid(kernel_size): + """Generate the mesh grid, centering at zero. + Args: + kernel_size (int): + Returns: + xy (ndarray): with the shape (kernel_size, kernel_size, 2) + xx (ndarray): with the shape (kernel_size, kernel_size) + yy (ndarray): with the shape (kernel_size, kernel_size) + """ + ax = np.arange(-kernel_size // 2 + 1., kernel_size // 2 + 1.) + xx, yy = np.meshgrid(ax, ax) + xy = np.hstack((xx.reshape((kernel_size * kernel_size, 1)), yy.reshape(kernel_size * kernel_size, + 1))).reshape(kernel_size, kernel_size, 2) + return xy, xx, yy + + +def pdf2(sigma_matrix, grid): + """Calculate PDF of the bivariate Gaussian distribution. + Args: + sigma_matrix (ndarray): with the shape (2, 2) + grid (ndarray): generated by :func:`mesh_grid`, + with the shape (K, K, 2), K is the kernel size. + Returns: + kernel (ndarrray): un-normalized kernel. + """ + inverse_sigma = np.linalg.inv(sigma_matrix) + kernel = np.exp(-0.5 * np.sum(np.dot(grid, inverse_sigma) * grid, 2)) + return kernel + +def bivariate_Gaussian(kernel_size, sig_x, sig_y, theta, grid=None, isotropic=True): + """Generate a bivariate isotropic or anisotropic Gaussian kernel. + In the isotropic mode, only `sig_x` is used. `sig_y` and `theta` is ignored. + Args: + kernel_size (int): + sig_x (float): + sig_y (float): + theta (float): Radian measurement. + grid (ndarray, optional): generated by :func:`mesh_grid`, + with the shape (K, K, 2), K is the kernel size. Default: None + isotropic (bool): + Returns: + kernel (ndarray): normalized kernel. + """ + if grid is None: + grid, _, _ = mesh_grid(kernel_size) + if isotropic: + sigma_matrix = np.array([[sig_x**2, 0], [0, sig_x**2]]) + else: + sigma_matrix = sigma_matrix2(sig_x, sig_y, theta) + kernel = pdf2(sigma_matrix, grid) + kernel = kernel / np.sum(kernel) + return kernel diff --git a/gradio_utils/traj_utils.py b/gradio_utils/traj_utils.py new file mode 100644 index 0000000000000000000000000000000000000000..bb0513c915a675c89347c1e461b5b8c461003155 --- /dev/null +++ b/gradio_utils/traj_utils.py @@ -0,0 +1,104 @@ +import cv2 +import numpy as np + +from gradio_utils.flow_utils import bivariate_Gaussian + +OBJECT_MOTION_MODE = ["Provided Trajectory", "Custom Trajectory"] + +PROVIDED_TRAJS = { + "horizon_1": "examples/trajectories/horizon_2.txt", + "swaying_1": "examples/trajectories/shake_1.txt", + "swaying_2": "examples/trajectories/shake_2.txt", + "swaying_3": "examples/trajectories/shaking_10.txt", + "curve_1": "examples/trajectories/curve_1.txt", + "curve_2": "examples/trajectories/curve_2.txt", + "curve_3": "examples/trajectories/curve_3.txt", + "curve_4": "examples/trajectories/curve_4.txt", +} + + +def read_points(file, video_len=16, reverse=False): + with open(file, 'r') as f: + lines = f.readlines() + points = [] + for line in lines: + x, y = line.strip().split(',') + points.append((int(x), int(y))) + if reverse: + points = points[::-1] + + if len(points) > video_len: + skip = len(points) // video_len + points = points[::skip] + points = points[:video_len] + + return points + +def get_provided_traj(traj_name): + traj = read_points(PROVIDED_TRAJS[traj_name]) + # xrange from 256 to 1024 + traj = [[int(1024*x/256), int(1024*y/256)] for x,y in traj] + return traj + +blur_kernel = bivariate_Gaussian(99, 10, 10, 0, grid=None, isotropic=True) + +def process_points(points): + frames = 16 + defualt_points = [[512,512]]*16 + + if len(points) < 2: + return defualt_points + elif len(points) >= frames: + skip = len(points)//frames + return points[::skip][:15] + points[-1:] + else: + insert_num = frames - len(points) + insert_num_dict = {} + interval = len(points) - 1 + n = insert_num // interval + m = insert_num % interval + for i in range(interval): + insert_num_dict[i] = n + for i in range(m): + insert_num_dict[i] += 1 + + res = [] + for i in range(interval): + insert_points = [] + x0,y0 = points[i] + x1,y1 = points[i+1] + + delta_x = x1 - x0 + delta_y = y1 - y0 + for j in range(insert_num_dict[i]): + x = x0 + (j+1)/(insert_num_dict[i]+1)*delta_x + y = y0 + (j+1)/(insert_num_dict[i]+1)*delta_y + insert_points.append([int(x), int(y)]) + + res += points[i:i+1] + insert_points + res += points[-1:] + return res + +def get_flow(points, video_len=16): + optical_flow = np.zeros((video_len, 256, 256, 2), dtype=np.float32) + for i in range(video_len-1): + p = points[i] + p1 = points[i+1] + optical_flow[i+1, p[1], p[0], 0] = p1[0] - p[0] + optical_flow[i+1, p[1], p[0], 1] = p1[1] - p[1] + for i in range(1, video_len): + optical_flow[i] = cv2.filter2D(optical_flow[i], -1, blur_kernel) + + + return optical_flow + + +def process_traj(points, device='cpu'): + xy_range = 1024 + points = process_points(points) + points = [[int(256*x/xy_range), int(256*y/xy_range)] for x,y in points] + + optical_flow = get_flow(points) + # optical_flow = torch.tensor(optical_flow).to(device) + + return optical_flow \ No newline at end of file diff --git a/gradio_utils/utils.py b/gradio_utils/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..9d783d0a7ed9791e7ff2c4f002282f8061e00348 --- /dev/null +++ b/gradio_utils/utils.py @@ -0,0 +1,175 @@ +import numpy as np +import plotly.express as px +import plotly.graph_objects as go + +def vis_camera(RT_list, rescale_T=1): + fig = go.Figure() + showticklabels = True + visible = True + scene_bounds = 2 + base_radius = 2.5 + zoom_scale = 1.5 + fov_deg = 50.0 + + edges = [(0, 1), (0, 2), (0, 3), (1, 2), (2, 3), (3, 1), (3, 4)] + + colors = px.colors.qualitative.Plotly + + cone_list = [] + n = len(RT_list) + for i, RT in enumerate(RT_list): + R = RT[:,:3] + T = RT[:,-1]/rescale_T + cone = calc_cam_cone_pts_3d(R, T, fov_deg) + cone_list.append((cone, (i*1/n, "green"), f"view_{i}")) + + + for (cone, clr, legend) in cone_list: + for (i, edge) in enumerate(edges): + (x1, x2) = (cone[edge[0], 0], cone[edge[1], 0]) + (y1, y2) = (cone[edge[0], 1], cone[edge[1], 1]) + (z1, z2) = (cone[edge[0], 2], cone[edge[1], 2]) + fig.add_trace(go.Scatter3d( + x=[x1, x2], y=[y1, y2], z=[z1, z2], mode='lines', + line=dict(color=clr, width=3), + name=legend, showlegend=(i == 0))) + fig.update_layout( + height=500, + autosize=True, + # hovermode=False, + margin=go.layout.Margin(l=0, r=0, b=0, t=0), + + showlegend=True, + legend=dict( + yanchor='bottom', + y=0.01, + xanchor='right', + x=0.99, + ), + scene=dict( + aspectmode='manual', + aspectratio=dict(x=1, y=1, z=1.0), + camera=dict( + center=dict(x=0.0, y=0.0, z=0.0), + up=dict(x=0.0, y=-1.0, z=0.0), + eye=dict(x=scene_bounds/2, y=-scene_bounds/2, z=-scene_bounds/2), + ), + + xaxis=dict( + range=[-scene_bounds, scene_bounds], + showticklabels=showticklabels, + visible=visible, + ), + + + yaxis=dict( + range=[-scene_bounds, scene_bounds], + showticklabels=showticklabels, + visible=visible, + ), + + + zaxis=dict( + range=[-scene_bounds, scene_bounds], + showticklabels=showticklabels, + visible=visible, + ) + )) + return fig + + +def calc_cam_cone_pts_3d(R_W2C, T_W2C, fov_deg, scale=0.1, set_canonical=False, first_frame_RT=None): + fov_rad = np.deg2rad(fov_deg) + R_W2C_inv = np.linalg.inv(R_W2C) + + # Camera pose center: + T = np.zeros_like(T_W2C) - T_W2C + T = np.dot(R_W2C_inv, T) + cam_x = T[0] + cam_y = T[1] + cam_z = T[2] + if set_canonical: + T = np.zeros_like(T_W2C) + T = np.dot(first_frame_RT[:,:3], T) + first_frame_RT[:,-1] + T = T - T_W2C + T = np.dot(R_W2C_inv, T) + cam_x = T[0] + cam_y = T[1] + cam_z = T[2] + + # vertex + corn1 = np.array([np.tan(fov_rad / 2.0), 0.5*np.tan(fov_rad / 2.0), 1.0]) *scale + corn2 = np.array([-np.tan(fov_rad / 2.0), 0.5*np.tan(fov_rad / 2.0), 1.0]) *scale + corn3 = np.array([0, -0.25*np.tan(fov_rad / 2.0), 1.0]) *scale + corn4 = np.array([0, -0.5*np.tan(fov_rad / 2.0), 1.0]) *scale + + corn1 = corn1 - T_W2C + corn2 = corn2 - T_W2C + corn3 = corn3 - T_W2C + corn4 = corn4 - T_W2C + + corn1 = np.dot(R_W2C_inv, corn1) + corn2 = np.dot(R_W2C_inv, corn2) + corn3 = np.dot(R_W2C_inv, corn3) + corn4 = np.dot(R_W2C_inv, corn4) + + # Now attach as offset to actual 3D camera position: + corn_x1 = corn1[0] + corn_y1 = corn1[1] + corn_z1 = corn1[2] + + corn_x2 = corn2[0] + corn_y2 = corn2[1] + corn_z2 = corn2[2] + + corn_x3 = corn3[0] + corn_y3 = corn3[1] + corn_z3 = corn3[2] + + corn_x4 = corn4[0] + corn_y4 = corn4[1] + corn_z4 = corn4[2] + + + xs = [cam_x, corn_x1, corn_x2, corn_x3, corn_x4, ] + ys = [cam_y, corn_y1, corn_y2, corn_y3, corn_y4, ] + zs = [cam_z, corn_z1, corn_z2, corn_z3, corn_z4, ] + + return np.array([xs, ys, zs]).T + + + + # T_base = [ + # [1.,0.,0.], ## W2C x 的正方向： 相机朝左 left + # [-1.,0.,0.], ## W2C x 的负方向： 相机朝右 right + # [0., 1., 0.], ## W2C y 的正方向： 相机朝上 up + # [0.,-1.,0.], ## W2C y 的负方向： 相机朝下 down + # [0.,0.,1.], ## W2C z 的正方向： 相机往前 zoom out + # [0.,0.,-1.], ## W2C z 的负方向： 相机往前 zoom in + # ] + # radius = 1 + # n = 16 + # # step = + # look_at = np.array([0, 0, 0.8]).reshape(3,1) + # # look_at = np.array([0, 0, 0.2]).reshape(3,1) + + # T_list = [] + # base_R = np.array([[1., 0., 0.], + # [0., 1., 0.], + # [0., 0., 1.]]) + # res = [] + # res_forsave = [] + # T_range = 1.8 + + + + # for i in range(0, 16): + # # theta = (1)*np.pi*i/n + + # R = base_R[:,:3] + # T = np.array([0.,0.,1.]).reshape(3,1) * (i/n)*2 + # RT = np.concatenate([R,T], axis=1) + # res.append(RT) + + # fig = vis_camera(res) + \ No newline at end of file diff --git a/lvdm/basics.py b/lvdm/basics.py new file mode 100644 index 0000000000000000000000000000000000000000..b310f8e2ecd0401bc2446cd6fa49e78e65d96dcf --- /dev/null +++ b/lvdm/basics.py @@ -0,0 +1,106 @@ +# adopted from +# https://github.com/openai/improved-diffusion/blob/main/improved_diffusion/gaussian_diffusion.py +# and +# https://github.com/lucidrains/denoising-diffusion-pytorch/blob/7706bdfc6f527f58d33f84b7b522e61e6e3164b3/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py +# and +# https://github.com/openai/guided-diffusion/blob/0ba878e517b276c45d1195eb29f6f5f72659a05b/guided_diffusion/nn.py +# +# thanks! + +import numpy as np +from einops import repeat + +import torch +import torch.nn as nn +import torch.nn.functional as F + +from utils.utils import instantiate_from_config + + +def disabled_train(self, mode=True): + """Overwrite model.train with this function to make sure train/eval mode + does not change anymore.""" + return self + +def zero_module(module): + """ + Zero out the parameters of a module and return it. + """ + for p in module.parameters(): + p.detach().zero_() + return module + +def scale_module(module, scale): + """ + Scale the parameters of a module and return it. + """ + for p in module.parameters(): + p.detach().mul_(scale) + return module + + +def conv_nd(dims, *args, **kwargs): + """ + Create a 1D, 2D, or 3D convolution module. + """ + if dims == 1: + return nn.Conv1d(*args, **kwargs) + elif dims == 2: + return nn.Conv2d(*args, **kwargs) + elif dims == 3: + return nn.Conv3d(*args, **kwargs) + raise ValueError(f"unsupported dimensions: {dims}") + + +def linear(*args, **kwargs): + """ + Create a linear module. + """ + return nn.Linear(*args, **kwargs) + + +def avg_pool_nd(dims, *args, **kwargs): + """ + Create a 1D, 2D, or 3D average pooling module. + """ + if dims == 1: + return nn.AvgPool1d(*args, **kwargs) + elif dims == 2: + return nn.AvgPool2d(*args, **kwargs) + elif dims == 3: + return nn.AvgPool3d(*args, **kwargs) + raise ValueError(f"unsupported dimensions: {dims}") + + +def nonlinearity(type='silu'): + if type == 'silu': + return nn.SiLU() + elif type == 'leaky_relu': + return nn.LeakyReLU() + + +class GroupNormSpecific(nn.GroupNorm): + def forward(self, x): + return super().forward(x.float()).type(x.dtype) + + +def normalization(channels, num_groups=32): + """ + Make a standard normalization layer. + :param channels: number of input channels. + :return: an nn.Module for normalization. + """ + return GroupNormSpecific(num_groups, channels) + + +class HybridConditioner(nn.Module): + + def __init__(self, c_concat_config, c_crossattn_config): + super().__init__() + self.concat_conditioner = instantiate_from_config(c_concat_config) + self.crossattn_conditioner = instantiate_from_config(c_crossattn_config) + + def forward(self, c_concat, c_crossattn): + c_concat = self.concat_conditioner(c_concat) + c_crossattn = self.crossattn_conditioner(c_crossattn) + return {'c_concat': [c_concat], 'c_crossattn': [c_crossattn]} \ No newline at end of file diff --git a/lvdm/common.py b/lvdm/common.py new file mode 100644 index 0000000000000000000000000000000000000000..d2cdee9dfe041780e22cdf6f2fcaf1aad6f4084e --- /dev/null +++ b/lvdm/common.py @@ -0,0 +1,142 @@ +import os, math +import numpy as np +from inspect import isfunction + +import torch +from torch import nn +import torch.nn.functional as F +import torch.distributed as dist + + +def gather_data(data, return_np=True): + ''' gather data from multiple processes to one list ''' + data_list = [torch.zeros_like(data) for _ in range(dist.get_world_size())] + dist.all_gather(data_list, data) # gather not supported with NCCL + if return_np: + data_list = [data.cpu().numpy() for data in data_list] + return data_list + +def autocast(f): + def do_autocast(*args, **kwargs): + with torch.cuda.amp.autocast(enabled=True, + dtype=torch.get_autocast_gpu_dtype(), + cache_enabled=torch.is_autocast_cache_enabled()): + return f(*args, **kwargs) + return do_autocast + + +def extract_into_tensor(a, t, x_shape): + b, *_ = t.shape + out = a.gather(-1, t) + return out.reshape(b, *((1,) * (len(x_shape) - 1))) + + +def noise_like(shape, device, repeat=False): + repeat_noise = lambda: torch.randn((1, *shape[1:]), device=device).repeat(shape[0], *((1,) * (len(shape) - 1))) + noise = lambda: torch.randn(shape, device=device) + return repeat_noise() if repeat else noise() + + +def default(val, d): + if exists(val): + return val + return d() if isfunction(d) else d + +def exists(val): + return val is not None + +def identity(*args, **kwargs): + return nn.Identity() + +def uniq(arr): + return{el: True for el in arr}.keys() + +def mean_flat(tensor): + """ + Take the mean over all non-batch dimensions. + """ + return tensor.mean(dim=list(range(1, len(tensor.shape)))) + +def ismap(x): + if not isinstance(x, torch.Tensor): + return False + return (len(x.shape) == 4) and (x.shape[1] > 3) + +def isimage(x): + if not isinstance(x,torch.Tensor): + return False + return (len(x.shape) == 4) and (x.shape[1] == 3 or x.shape[1] == 1) + +def max_neg_value(t): + return -torch.finfo(t.dtype).max + +def shape_to_str(x): + shape_str = "x".join([str(x) for x in x.shape]) + return shape_str + +def init_(tensor): + dim = tensor.shape[-1] + std = 1 / math.sqrt(dim) + tensor.uniform_(-std, std) + return tensor + +#import deepspeed +#ckpt = deepspeed.checkpointing.checkpoint +ckpt = torch.utils.checkpoint.checkpoint +def checkpoint(func, inputs, params, flag): + """ + Evaluate a function without caching intermediate activations, allowing for + reduced memory at the expense of extra compute in the backward pass. + :param func: the function to evaluate. + :param inputs: the argument sequence to pass to `func`. + :param params: a sequence of parameters `func` depends on but does not + explicitly take as arguments. + :param flag: if False, disable gradient checkpointing. + """ + if flag: + try: + return ckpt(func, *inputs) + except: + args = tuple(inputs) + tuple(params) + return CheckpointFunction.apply(func, len(inputs), *args) + else: + return func(*inputs) + + +class CheckpointFunction(torch.autograd.Function): + @staticmethod + @torch.cuda.amp.custom_fwd + def forward(ctx, run_function, length, *args): + ctx.run_function = run_function + ctx.input_tensors = list(args[:length]) + ctx.input_params = list(args[length:]) + + with torch.no_grad(): + output_tensors = ctx.run_function(*ctx.input_tensors) + return output_tensors + + @staticmethod + @torch.cuda.amp.custom_bwd # add this + def backward(ctx, *output_grads): + ''' + for x in ctx.input_tensors: + if isinstance(x, int): + print('-----------------', ctx.run_function) + ''' + ctx.input_tensors = [x.detach().requires_grad_(True) for x in ctx.input_tensors] + with torch.enable_grad(): + # Fixes a bug where the first op in run_function modifies the + # Tensor storage in place, which is not allowed for detach()'d + # Tensors. + shallow_copies = [x.view_as(x) for x in ctx.input_tensors] + output_tensors = ctx.run_function(*shallow_copies) + input_grads = torch.autograd.grad( + output_tensors, + ctx.input_tensors + ctx.input_params, + output_grads, + allow_unused=True, + ) + del ctx.input_tensors + del ctx.input_params + del output_tensors + return (None, None) + input_grads diff --git a/lvdm/distributions.py b/lvdm/distributions.py new file mode 100644 index 0000000000000000000000000000000000000000..0b69b6984880ec24279b658384ed8031335e3474 --- /dev/null +++ b/lvdm/distributions.py @@ -0,0 +1,95 @@ +import torch +import numpy as np + + +class AbstractDistribution: + def sample(self): + raise NotImplementedError() + + def mode(self): + raise NotImplementedError() + + +class DiracDistribution(AbstractDistribution): + def __init__(self, value): + self.value = value + + def sample(self): + return self.value + + def mode(self): + return self.value + + +class DiagonalGaussianDistribution(object): + def __init__(self, parameters, deterministic=False): + self.parameters = parameters + self.mean, self.logvar = torch.chunk(parameters, 2, dim=1) + self.logvar = torch.clamp(self.logvar, -30.0, 20.0) + self.deterministic = deterministic + self.std = torch.exp(0.5 * self.logvar) + self.var = torch.exp(self.logvar) + if self.deterministic: + self.var = self.std = torch.zeros_like(self.mean).to(device=self.parameters.device) + + def sample(self, noise=None): + if noise is None: + noise = torch.randn(self.mean.shape) + + x = self.mean + self.std * noise.to(device=self.parameters.device) + return x + + def kl(self, other=None): + if self.deterministic: + return torch.Tensor([0.]) + else: + if other is None: + return 0.5 * torch.sum(torch.pow(self.mean, 2) + + self.var - 1.0 - self.logvar, + dim=[1, 2, 3]) + else: + return 0.5 * torch.sum( + torch.pow(self.mean - other.mean, 2) / other.var + + self.var / other.var - 1.0 - self.logvar + other.logvar, + dim=[1, 2, 3]) + + def nll(self, sample, dims=[1,2,3]): + if self.deterministic: + return torch.Tensor([0.]) + logtwopi = np.log(2.0 * np.pi) + return 0.5 * torch.sum( + logtwopi + self.logvar + torch.pow(sample - self.mean, 2) / self.var, + dim=dims) + + def mode(self): + return self.mean + + +def normal_kl(mean1, logvar1, mean2, logvar2): + """ + source: https://github.com/openai/guided-diffusion/blob/27c20a8fab9cb472df5d6bdd6c8d11c8f430b924/guided_diffusion/losses.py#L12 + Compute the KL divergence between two gaussians. + Shapes are automatically broadcasted, so batches can be compared to + scalars, among other use cases. + """ + tensor = None + for obj in (mean1, logvar1, mean2, logvar2): + if isinstance(obj, torch.Tensor): + tensor = obj + break + assert tensor is not None, "at least one argument must be a Tensor" + + # Force variances to be Tensors. Broadcasting helps convert scalars to + # Tensors, but it does not work for torch.exp(). + logvar1, logvar2 = [ + x if isinstance(x, torch.Tensor) else torch.tensor(x).to(tensor) + for x in (logvar1, logvar2) + ] + + return 0.5 * ( + -1.0 + + logvar2 + - logvar1 + + torch.exp(logvar1 - logvar2) + + ((mean1 - mean2) ** 2) * torch.exp(-logvar2) + ) diff --git a/lvdm/ema.py b/lvdm/ema.py new file mode 100644 index 0000000000000000000000000000000000000000..c8c75af43565f6e140287644aaaefa97dd6e67c5 --- /dev/null +++ b/lvdm/ema.py @@ -0,0 +1,76 @@ +import torch +from torch import nn + + +class LitEma(nn.Module): + def __init__(self, model, decay=0.9999, use_num_upates=True): + super().__init__() + if decay < 0.0 or decay > 1.0: + raise ValueError('Decay must be between 0 and 1') + + self.m_name2s_name = {} + self.register_buffer('decay', torch.tensor(decay, dtype=torch.float32)) + self.register_buffer('num_updates', torch.tensor(0,dtype=torch.int) if use_num_upates + else torch.tensor(-1,dtype=torch.int)) + + for name, p in model.named_parameters(): + if p.requires_grad: + #remove as '.'-character is not allowed in buffers + s_name = name.replace('.','') + self.m_name2s_name.update({name:s_name}) + self.register_buffer(s_name,p.clone().detach().data) + + self.collected_params = [] + + def forward(self,model): + decay = self.decay + + if self.num_updates >= 0: + self.num_updates += 1 + decay = min(self.decay,(1 + self.num_updates) / (10 + self.num_updates)) + + one_minus_decay = 1.0 - decay + + with torch.no_grad(): + m_param = dict(model.named_parameters()) + shadow_params = dict(self.named_buffers()) + + for key in m_param: + if m_param[key].requires_grad: + sname = self.m_name2s_name[key] + shadow_params[sname] = shadow_params[sname].type_as(m_param[key]) + shadow_params[sname].sub_(one_minus_decay * (shadow_params[sname] - m_param[key])) + else: + assert not key in self.m_name2s_name + + def copy_to(self, model): + m_param = dict(model.named_parameters()) + shadow_params = dict(self.named_buffers()) + for key in m_param: + if m_param[key].requires_grad: + m_param[key].data.copy_(shadow_params[self.m_name2s_name[key]].data) + else: + assert not key in self.m_name2s_name + + def store(self, parameters): + """ + Save the current parameters for restoring later. + Args: + parameters: Iterable of `torch.nn.Parameter`; the parameters to be + temporarily stored. + """ + self.collected_params = [param.clone() for param in parameters] + + def restore(self, parameters): + """ + Restore the parameters stored with the `store` method. + Useful to validate the model with EMA parameters without affecting the + original optimization process. Store the parameters before the + `copy_to` method. After validation (or model saving), use this to + restore the former parameters. + Args: + parameters: Iterable of `torch.nn.Parameter`; the parameters to be + updated with the stored parameters. + """ + for c_param, param in zip(self.collected_params, parameters): + param.data.copy_(c_param.data) diff --git a/lvdm/models/autoencoder.py b/lvdm/models/autoencoder.py new file mode 100644 index 0000000000000000000000000000000000000000..869f1b1651da605d9868acded3fdf3cc0e1c6f87 --- /dev/null +++ b/lvdm/models/autoencoder.py @@ -0,0 +1,221 @@ +import os +from contextlib import contextmanager + +import numpy as np +import pytorch_lightning as pl +import torch +import torch.nn.functional as F +from einops import rearrange + +from lvdm.distributions import DiagonalGaussianDistribution +from lvdm.modules.networks.ae_modules import Decoder, Encoder +from utils.utils import instantiate_from_config + + +class AutoencoderKL(pl.LightningModule): + def __init__(self, + ddconfig, + lossconfig, + embed_dim, + ckpt_path=None, + ignore_keys=[], + image_key="image", + colorize_nlabels=None, + monitor=None, + test=False, + logdir=None, + input_dim=4, + test_args=None, + ): + super().__init__() + self.image_key = image_key + self.encoder = Encoder(**ddconfig) + self.decoder = Decoder(**ddconfig) + self.loss = instantiate_from_config(lossconfig) + assert ddconfig["double_z"] + self.quant_conv = torch.nn.Conv2d(2*ddconfig["z_channels"], 2*embed_dim, 1) + self.post_quant_conv = torch.nn.Conv2d(embed_dim, ddconfig["z_channels"], 1) + self.embed_dim = embed_dim + self.input_dim = input_dim + self.test = test + self.test_args = test_args + self.logdir = logdir + if colorize_nlabels is not None: + assert type(colorize_nlabels)==int + self.register_buffer("colorize", torch.randn(3, colorize_nlabels, 1, 1)) + if monitor is not None: + self.monitor = monitor + if ckpt_path is not None: + self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys) + if self.test: + self.init_test() + + def init_test(self,): + self.test = True + save_dir = os.path.join(self.logdir, "test") + if 'ckpt' in self.test_args: + ckpt_name = os.path.basename(self.test_args.ckpt).split('.ckpt')[0] + f'_epoch{self._cur_epoch}' + self.root = os.path.join(save_dir, ckpt_name) + else: + self.root = save_dir + if 'test_subdir' in self.test_args: + self.root = os.path.join(save_dir, self.test_args.test_subdir) + + self.root_zs = os.path.join(self.root, "zs") + self.root_dec = os.path.join(self.root, "reconstructions") + self.root_inputs = os.path.join(self.root, "inputs") + os.makedirs(self.root, exist_ok=True) + + if self.test_args.save_z: + os.makedirs(self.root_zs, exist_ok=True) + if self.test_args.save_reconstruction: + os.makedirs(self.root_dec, exist_ok=True) + if self.test_args.save_input: + os.makedirs(self.root_inputs, exist_ok=True) + assert(self.test_args is not None) + self.test_maximum = getattr(self.test_args, 'test_maximum', None) + self.count = 0 + self.eval_metrics = {} + self.decodes = [] + self.save_decode_samples = 2048 + + def init_from_ckpt(self, path, ignore_keys=list()): + sd = torch.load(path, map_location="cpu") + try: + self._cur_epoch = sd['epoch'] + sd = sd["state_dict"] + except: + self._cur_epoch = 'null' + keys = list(sd.keys()) + for k in keys: + for ik in ignore_keys: + if k.startswith(ik): + print("Deleting key {} from state_dict.".format(k)) + del sd[k] + self.load_state_dict(sd, strict=False) + # self.load_state_dict(sd, strict=True) + print(f"Restored from {path}") + + def encode(self, x, **kwargs): + + h = self.encoder(x) + moments = self.quant_conv(h) + posterior = DiagonalGaussianDistribution(moments) + return posterior + + def decode(self, z, **kwargs): + z = self.post_quant_conv(z) + dec = self.decoder(z) + return dec + + def forward(self, input, sample_posterior=True): + posterior = self.encode(input) + if sample_posterior: + z = posterior.sample() + else: + z = posterior.mode() + dec = self.decode(z) + return dec, posterior + + def get_input(self, batch, k): + x = batch[k] + if x.dim() == 5 and self.input_dim == 4: + b,c,t,h,w = x.shape + self.b = b + self.t = t + x = rearrange(x, 'b c t h w -> (b t) c h w') + + return x + + def training_step(self, batch, batch_idx, optimizer_idx): + inputs = self.get_input(batch, self.image_key) + reconstructions, posterior = self(inputs) + + if optimizer_idx == 0: + # train encoder+decoder+logvar + aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step, + last_layer=self.get_last_layer(), split="train") + self.log("aeloss", aeloss, prog_bar=True, logger=True, on_step=True, on_epoch=True) + self.log_dict(log_dict_ae, prog_bar=False, logger=True, on_step=True, on_epoch=False) + return aeloss + + if optimizer_idx == 1: + # train the discriminator + discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, optimizer_idx, self.global_step, + last_layer=self.get_last_layer(), split="train") + + self.log("discloss", discloss, prog_bar=True, logger=True, on_step=True, on_epoch=True) + self.log_dict(log_dict_disc, prog_bar=False, logger=True, on_step=True, on_epoch=False) + return discloss + + def validation_step(self, batch, batch_idx): + inputs = self.get_input(batch, self.image_key) + reconstructions, posterior = self(inputs) + aeloss, log_dict_ae = self.loss(inputs, reconstructions, posterior, 0, self.global_step, + last_layer=self.get_last_layer(), split="val") + + discloss, log_dict_disc = self.loss(inputs, reconstructions, posterior, 1, self.global_step, + last_layer=self.get_last_layer(), split="val") + + self.log("val/rec_loss", log_dict_ae["val/rec_loss"]) + self.log_dict(log_dict_ae) + self.log_dict(log_dict_disc) + return self.log_dict + + def configure_optimizers(self): + lr = self.learning_rate + opt_ae = torch.optim.Adam(list(self.encoder.parameters())+ + list(self.decoder.parameters())+ + list(self.quant_conv.parameters())+ + list(self.post_quant_conv.parameters()), + lr=lr, betas=(0.5, 0.9)) + opt_disc = torch.optim.Adam(self.loss.discriminator.parameters(), + lr=lr, betas=(0.5, 0.9)) + return [opt_ae, opt_disc], [] + + def get_last_layer(self): + return self.decoder.conv_out.weight + + @torch.no_grad() + def log_images(self, batch, only_inputs=False, **kwargs): + log = dict() + x = self.get_input(batch, self.image_key) + x = x.to(self.device) + if not only_inputs: + xrec, posterior = self(x) + if x.shape[1] > 3: + # colorize with random projection + assert xrec.shape[1] > 3 + x = self.to_rgb(x) + xrec = self.to_rgb(xrec) + log["samples"] = self.decode(torch.randn_like(posterior.sample())) + log["reconstructions"] = xrec + log["inputs"] = x + return log + + def to_rgb(self, x): + assert self.image_key == "segmentation" + if not hasattr(self, "colorize"): + self.register_buffer("colorize", torch.randn(3, x.shape[1], 1, 1).to(x)) + x = F.conv2d(x, weight=self.colorize) + x = 2.*(x-x.min())/(x.max()-x.min()) - 1. + return x + +class IdentityFirstStage(torch.nn.Module): + def __init__(self, *args, vq_interface=False, **kwargs): + self.vq_interface = vq_interface # TODO: Should be true by default but check to not break older stuff + super().__init__() + + def encode(self, x, *args, **kwargs): + return x + + def decode(self, x, *args, **kwargs): + return x + + def quantize(self, x, *args, **kwargs): + if self.vq_interface: + return x, None, [None, None, None] + return x + + def forward(self, x, *args, **kwargs): + return x \ No newline at end of file diff --git a/lvdm/models/ddpm3d.py b/lvdm/models/ddpm3d.py new file mode 100644 index 0000000000000000000000000000000000000000..56c0b75235b83a9aab0670a41a5a9224a429f5fc --- /dev/null +++ b/lvdm/models/ddpm3d.py @@ -0,0 +1,1085 @@ +""" +wild mixture of +https://github.com/openai/improved-diffusion/blob/e94489283bb876ac1477d5dd7709bbbd2d9902ce/improved_diffusion/gaussian_diffusion.py +https://github.com/lucidrains/denoising-diffusion-pytorch/blob/7706bdfc6f527f58d33f84b7b522e61e6e3164b3/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py +https://github.com/CompVis/taming-transformers +-- merci +""" + +import logging +import os +import random +from contextlib import contextmanager +from functools import partial + +import numpy as np +from einops import rearrange, repeat +from tqdm import tqdm + +mainlogger = logging.getLogger('mainlogger') + +import pytorch_lightning as pl +import torch +import torch.nn as nn +from pytorch_lightning.utilities import rank_zero_only +from torch.optim.lr_scheduler import CosineAnnealingLR, LambdaLR +from torchvision.utils import make_grid + +from lvdm.basics import disabled_train +from lvdm.common import default, exists, extract_into_tensor, noise_like +from lvdm.distributions import DiagonalGaussianDistribution, normal_kl +from lvdm.ema import LitEma +from lvdm.models.samplers.ddim import DDIMSampler +from lvdm.models.utils_diffusion import make_beta_schedule +from utils.utils import instantiate_from_config + +__conditioning_keys__ = {'concat': 'c_concat', + 'crossattn': 'c_crossattn', + 'adm': 'y'} + +class DDPM(pl.LightningModule): + # classic DDPM with Gaussian diffusion, in image space + def __init__(self, + unet_config, + timesteps=1000, + beta_schedule="linear", + loss_type="l2", + ckpt_path=None, + ignore_keys=[], + load_only_unet=False, + monitor=None, + use_ema=True, + first_stage_key="image", + image_size=256, + channels=3, + log_every_t=100, + clip_denoised=True, + linear_start=1e-4, + linear_end=2e-2, + cosine_s=8e-3, + given_betas=None, + original_elbo_weight=0., + v_posterior=0., # weight for choosing posterior variance as sigma = (1-v) * beta_tilde + v * beta + l_simple_weight=1., + conditioning_key=None, + parameterization="eps", # all assuming fixed variance schedules + scheduler_config=None, + use_positional_encodings=False, + learn_logvar=False, + logvar_init=0., + ): + super().__init__() + assert parameterization in ["eps", "x0"], 'currently only supporting "eps" and "x0"' + self.parameterization = parameterization + mainlogger.info(f"{self.__class__.__name__}: Running in {self.parameterization}-prediction mode") + self.cond_stage_model = None + self.clip_denoised = clip_denoised + self.log_every_t = log_every_t + self.first_stage_key = first_stage_key + self.channels = channels + self.temporal_length = unet_config.params.temporal_length + self.image_size = image_size # try conv? + if isinstance(self.image_size, int): + self.image_size = [self.image_size, self.image_size] + self.use_positional_encodings = use_positional_encodings + self.model = DiffusionWrapper(unet_config, conditioning_key) + #count_params(self.model, verbose=True) + self.use_ema = use_ema + if self.use_ema: + self.model_ema = LitEma(self.model) + mainlogger.info(f"Keeping EMAs of {len(list(self.model_ema.buffers()))}.") + + self.use_scheduler = scheduler_config is not None + if self.use_scheduler: + self.scheduler_config = scheduler_config + + self.v_posterior = v_posterior + self.original_elbo_weight = original_elbo_weight + self.l_simple_weight = l_simple_weight + + if monitor is not None: + self.monitor = monitor + if ckpt_path is not None: + self.init_from_ckpt(ckpt_path, ignore_keys=ignore_keys, only_model=load_only_unet) + + self.register_schedule(given_betas=given_betas, beta_schedule=beta_schedule, timesteps=timesteps, + linear_start=linear_start, linear_end=linear_end, cosine_s=cosine_s) + + self.loss_type = loss_type + + self.learn_logvar = learn_logvar + self.logvar = torch.full(fill_value=logvar_init, size=(self.num_timesteps,)) + if self.learn_logvar: + self.logvar = nn.Parameter(self.logvar, requires_grad=True) + + def register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000, + linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3): + if exists(given_betas): + betas = given_betas + else: + betas = make_beta_schedule(beta_schedule, timesteps, linear_start=linear_start, linear_end=linear_end, + cosine_s=cosine_s) + alphas = 1. - betas + alphas_cumprod = np.cumprod(alphas, axis=0) + alphas_cumprod_prev = np.append(1., alphas_cumprod[:-1]) + + timesteps, = betas.shape + self.num_timesteps = int(timesteps) + self.linear_start = linear_start + self.linear_end = linear_end + assert alphas_cumprod.shape[0] == self.num_timesteps, 'alphas have to be defined for each timestep' + + to_torch = partial(torch.tensor, dtype=torch.float32) + + self.register_buffer('betas', to_torch(betas)) + self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod)) + self.register_buffer('alphas_cumprod_prev', to_torch(alphas_cumprod_prev)) + + # calculations for diffusion q(x_t | x_{t-1}) and others + self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod))) + self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod))) + self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod))) + self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod))) + self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod - 1))) + + # calculations for posterior q(x_{t-1} | x_t, x_0) + posterior_variance = (1 - self.v_posterior) * betas * (1. - alphas_cumprod_prev) / ( + 1. - alphas_cumprod) + self.v_posterior * betas + # above: equal to 1. / (1. / (1. - alpha_cumprod_tm1) + alpha_t / beta_t) + self.register_buffer('posterior_variance', to_torch(posterior_variance)) + # below: log calculation clipped because the posterior variance is 0 at the beginning of the diffusion chain + self.register_buffer('posterior_log_variance_clipped', to_torch(np.log(np.maximum(posterior_variance, 1e-20)))) + self.register_buffer('posterior_mean_coef1', to_torch( + betas * np.sqrt(alphas_cumprod_prev) / (1. - alphas_cumprod))) + self.register_buffer('posterior_mean_coef2', to_torch( + (1. - alphas_cumprod_prev) * np.sqrt(alphas) / (1. - alphas_cumprod))) + + if self.parameterization == "eps": + lvlb_weights = self.betas ** 2 / ( + 2 * self.posterior_variance * to_torch(alphas) * (1 - self.alphas_cumprod)) + elif self.parameterization == "x0": + lvlb_weights = 0.5 * np.sqrt(torch.Tensor(alphas_cumprod)) / (2. * 1 - torch.Tensor(alphas_cumprod)) + else: + raise NotImplementedError("mu not supported") + # TODO how to choose this term + lvlb_weights[0] = lvlb_weights[1] + self.register_buffer('lvlb_weights', lvlb_weights, persistent=False) + assert not torch.isnan(self.lvlb_weights).all() + + @contextmanager + def ema_scope(self, context=None): + if self.use_ema: + self.model_ema.store(self.model.parameters()) + self.model_ema.copy_to(self.model) + if context is not None: + mainlogger.info(f"{context}: Switched to EMA weights") + try: + yield None + finally: + if self.use_ema: + self.model_ema.restore(self.model.parameters()) + if context is not None: + mainlogger.info(f"{context}: Restored training weights") + + def init_from_ckpt(self, path, ignore_keys=list(), only_model=False): + sd = torch.load(path, map_location="cpu") + if "state_dict" in list(sd.keys()): + sd = sd["state_dict"] + keys = list(sd.keys()) + for k in keys: + for ik in ignore_keys: + if k.startswith(ik): + mainlogger.info("Deleting key {} from state_dict.".format(k)) + del sd[k] + missing, unexpected = self.load_state_dict(sd, strict=False) if not only_model else self.model.load_state_dict( + sd, strict=False) + mainlogger.info(f"Restored from {path} with {len(missing)} missing and {len(unexpected)} unexpected keys") + if len(missing) > 0: + mainlogger.info(f"Missing Keys: {missing}") + if len(unexpected) > 0: + mainlogger.info(f"Unexpected Keys: {unexpected}") + + def q_mean_variance(self, x_start, t): + """ + Get the distribution q(x_t | x_0). + :param x_start: the [N x C x ...] tensor of noiseless inputs. + :param t: the number of diffusion steps (minus 1). Here, 0 means one step. + :return: A tuple (mean, variance, log_variance), all of x_start's shape. + """ + mean = (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start) + variance = extract_into_tensor(1.0 - self.alphas_cumprod, t, x_start.shape) + log_variance = extract_into_tensor(self.log_one_minus_alphas_cumprod, t, x_start.shape) + return mean, variance, log_variance + + def predict_start_from_noise(self, x_t, t, noise): + return ( + extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - + extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * noise + ) + + def q_posterior(self, x_start, x_t, t): + posterior_mean = ( + extract_into_tensor(self.posterior_mean_coef1, t, x_t.shape) * x_start + + extract_into_tensor(self.posterior_mean_coef2, t, x_t.shape) * x_t + ) + posterior_variance = extract_into_tensor(self.posterior_variance, t, x_t.shape) + posterior_log_variance_clipped = extract_into_tensor(self.posterior_log_variance_clipped, t, x_t.shape) + return posterior_mean, posterior_variance, posterior_log_variance_clipped + + def p_mean_variance(self, x, t, clip_denoised: bool): + model_out = self.model(x, t) + if self.parameterization == "eps": + x_recon = self.predict_start_from_noise(x, t=t, noise=model_out) + elif self.parameterization == "x0": + x_recon = model_out + if clip_denoised: + x_recon.clamp_(-1., 1.) + + model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t) + return model_mean, posterior_variance, posterior_log_variance + + @torch.no_grad() + def p_sample(self, x, t, clip_denoised=True, repeat_noise=False): + b, *_, device = *x.shape, x.device + model_mean, _, model_log_variance = self.p_mean_variance(x=x, t=t, clip_denoised=clip_denoised) + noise = noise_like(x.shape, device, repeat_noise) + # no noise when t == 0 + nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1))) + return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise + + @torch.no_grad() + def p_sample_loop(self, shape, return_intermediates=False): + device = self.betas.device + b = shape[0] + img = torch.randn(shape, device=device) + intermediates = [img] + for i in tqdm(reversed(range(0, self.num_timesteps)), desc='Sampling t', total=self.num_timesteps): + img = self.p_sample(img, torch.full((b,), i, device=device, dtype=torch.long), + clip_denoised=self.clip_denoised) + if i % self.log_every_t == 0 or i == self.num_timesteps - 1: + intermediates.append(img) + if return_intermediates: + return img, intermediates + return img + + @torch.no_grad() + def sample(self, batch_size=16, return_intermediates=False): + image_size = self.image_size + channels = self.channels + return self.p_sample_loop((batch_size, channels, image_size, image_size), + return_intermediates=return_intermediates) + + def q_sample(self, x_start, t, noise=None): + noise = default(noise, lambda: torch.randn_like(x_start)) + return (extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start + + extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise) + + def get_loss(self, pred, target, mean=True): + if self.loss_type == 'l1': + loss = (target - pred).abs() + if mean: + loss = loss.mean() + elif self.loss_type == 'l2': + if mean: + loss = torch.nn.functional.mse_loss(target, pred) + else: + loss = torch.nn.functional.mse_loss(target, pred, reduction='none') + else: + raise NotImplementedError("unknown loss type '{loss_type}'") + + return loss + + def p_losses(self, x_start, t, noise=None): + noise = default(noise, lambda: torch.randn_like(x_start)) + x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise) + model_out = self.model(x_noisy, t) + + loss_dict = {} + if self.parameterization == "eps": + target = noise + elif self.parameterization == "x0": + target = x_start + else: + raise NotImplementedError(f"Paramterization {self.parameterization} not yet supported") + + loss = self.get_loss(model_out, target, mean=False).mean(dim=[1, 2, 3]) + + log_prefix = 'train' if self.training else 'val' + + loss_dict.update({f'{log_prefix}/loss_simple': loss.mean()}) + loss_simple = loss.mean() * self.l_simple_weight + + loss_vlb = (self.lvlb_weights[t] * loss).mean() + loss_dict.update({f'{log_prefix}/loss_vlb': loss_vlb}) + + loss = loss_simple + self.original_elbo_weight * loss_vlb + + loss_dict.update({f'{log_prefix}/loss': loss}) + + return loss, loss_dict + + def forward(self, x, *args, **kwargs): + # b, c, h, w, device, img_size, = *x.shape, x.device, self.image_size + # assert h == img_size and w == img_size, f'height and width of image must be {img_size}' + t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long() + return self.p_losses(x, t, *args, **kwargs) + + def get_input(self, batch, k): + x = batch[k] + ''' + if len(x.shape) == 3: + x = x[..., None] + x = rearrange(x, 'b h w c -> b c h w') + ''' + x = x.to(memory_format=torch.contiguous_format).float() + return x + + def shared_step(self, batch): + x = self.get_input(batch, self.first_stage_key) + loss, loss_dict = self(x) + return loss, loss_dict + + def training_step(self, batch, batch_idx): + loss, loss_dict = self.shared_step(batch) + + self.log_dict(loss_dict, prog_bar=True, + logger=True, on_step=True, on_epoch=True) + + self.log("global_step", self.global_step, + prog_bar=True, logger=True, on_step=True, on_epoch=False) + + if self.use_scheduler: + lr = self.optimizers().param_groups[0]['lr'] + self.log('lr_abs', lr, prog_bar=True, logger=True, on_step=True, on_epoch=False) + + return loss + + @torch.no_grad() + def validation_step(self, batch, batch_idx): + _, loss_dict_no_ema = self.shared_step(batch) + with self.ema_scope(): + _, loss_dict_ema = self.shared_step(batch) + loss_dict_ema = {key + '_ema': loss_dict_ema[key] for key in loss_dict_ema} + self.log_dict(loss_dict_no_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True) + self.log_dict(loss_dict_ema, prog_bar=False, logger=True, on_step=False, on_epoch=True) + + def on_train_batch_end(self, *args, **kwargs): + if self.use_ema: + self.model_ema(self.model) + + def _get_rows_from_list(self, samples): + n_imgs_per_row = len(samples) + denoise_grid = rearrange(samples, 'n b c h w -> b n c h w') + denoise_grid = rearrange(denoise_grid, 'b n c h w -> (b n) c h w') + denoise_grid = make_grid(denoise_grid, nrow=n_imgs_per_row) + return denoise_grid + + @torch.no_grad() + def log_images(self, batch, N=8, n_row=2, sample=True, return_keys=None, **kwargs): + log = dict() + x = self.get_input(batch, self.first_stage_key) + N = min(x.shape[0], N) + n_row = min(x.shape[0], n_row) + x = x.to(self.device)[:N] + log["inputs"] = x + + # get diffusion row + diffusion_row = list() + x_start = x[:n_row] + + for t in range(self.num_timesteps): + if t % self.log_every_t == 0 or t == self.num_timesteps - 1: + t = repeat(torch.tensor([t]), '1 -> b', b=n_row) + t = t.to(self.device).long() + noise = torch.randn_like(x_start) + x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise) + diffusion_row.append(x_noisy) + + log["diffusion_row"] = self._get_rows_from_list(diffusion_row) + + if sample: + # get denoise row + with self.ema_scope("Plotting"): + samples, denoise_row = self.sample(batch_size=N, return_intermediates=True) + + log["samples"] = samples + log["denoise_row"] = self._get_rows_from_list(denoise_row) + + if return_keys: + if np.intersect1d(list(log.keys()), return_keys).shape[0] == 0: + return log + else: + return {key: log[key] for key in return_keys} + return log + + def configure_optimizers(self): + lr = self.learning_rate + params = list(self.model.parameters()) + if self.learn_logvar: + params = params + [self.logvar] + opt = torch.optim.AdamW(params, lr=lr) + return opt + + +class LatentDiffusion(DDPM): + """main class""" + def __init__(self, + first_stage_config, + cond_stage_config, + num_timesteps_cond=None, + cond_stage_key="caption", + cond_stage_trainable=False, + cond_stage_forward=None, + conditioning_key=None, + uncond_prob=0.2, + uncond_type="empty_seq", + scale_factor=1.0, + scale_by_std=False, + # added for LVDM + encoder_type="2d", + frame_cond=None, + only_model=False, + logdir=None, + empty_params_only=False, + *args, **kwargs): + self.num_timesteps_cond = default(num_timesteps_cond, 1) + self.scale_by_std = scale_by_std + assert self.num_timesteps_cond <= kwargs['timesteps'] + # for backwards compatibility after implementation of DiffusionWrapper + ckpt_path = kwargs.pop("ckpt_path", None) + ignore_keys = kwargs.pop("ignore_keys", []) + conditioning_key = default(conditioning_key, 'crossattn') + super().__init__(conditioning_key=conditioning_key, *args, **kwargs) + + self.cond_stage_trainable = cond_stage_trainable + self.cond_stage_key = cond_stage_key + self.empty_params_only = empty_params_only + try: + self.num_downs = len(first_stage_config.params.ddconfig.ch_mult) - 1 + except: + self.num_downs = 0 + if not scale_by_std: + self.scale_factor = scale_factor + else: + self.register_buffer('scale_factor', torch.tensor(scale_factor)) + self.instantiate_first_stage(first_stage_config) + self.instantiate_cond_stage(cond_stage_config) + self.first_stage_config = first_stage_config + self.cond_stage_config = cond_stage_config + self.clip_denoised = False + + self.cond_stage_forward = cond_stage_forward + self.encoder_type = encoder_type + assert(encoder_type in ["2d", "3d"]) + self.uncond_prob = uncond_prob + self.classifier_free_guidance = True if uncond_prob > 0 else False + assert(uncond_type in ["zero_embed", "empty_seq"]) + self.uncond_type = uncond_type + + ## future frame prediction + self.frame_cond = frame_cond + if self.frame_cond: + # frame_len = self.model.diffusion_model.temporal_length + frame_len = self.temporal_length + cond_mask = torch.zeros(frame_len, dtype=torch.float32) + cond_mask[:self.frame_cond] = 1.0 + ## b,c,t,h,w + self.cond_mask = cond_mask[None,None,:,None,None] + mainlogger.info("---training for %d-frame conditoning T2V"%(self.frame_cond)) + else: + self.cond_mask = None + + self.restarted_from_ckpt = False + if ckpt_path is not None: + self.init_from_ckpt(ckpt_path, ignore_keys, only_model=only_model) + self.restarted_from_ckpt = True + + self.logdir = logdir + + def make_cond_schedule(self, ): + self.cond_ids = torch.full(size=(self.num_timesteps,), fill_value=self.num_timesteps - 1, dtype=torch.long) + ids = torch.round(torch.linspace(0, self.num_timesteps - 1, self.num_timesteps_cond)).long() + self.cond_ids[:self.num_timesteps_cond] = ids + + @rank_zero_only + @torch.no_grad() + def on_train_batch_start(self, batch, batch_idx, dataloader_idx=None): + # only for very first batch, reset the self.scale_factor + if self.scale_by_std and self.current_epoch == 0 and self.global_step == 0 and batch_idx == 0 and \ + not self.restarted_from_ckpt: + assert self.scale_factor == 1., 'rather not use custom rescaling and std-rescaling simultaneously' + # set rescale weight to 1./std of encodings + mainlogger.info("### USING STD-RESCALING ###") + x = super().get_input(batch, self.first_stage_key) + x = x.to(self.device) + encoder_posterior = self.encode_first_stage(x) + z = self.get_first_stage_encoding(encoder_posterior).detach() + del self.scale_factor + self.register_buffer('scale_factor', 1. / z.flatten().std()) + mainlogger.info(f"setting self.scale_factor to {self.scale_factor}") + mainlogger.info("### USING STD-RESCALING ###") + mainlogger.info(f"std={z.flatten().std()}") + + def register_schedule(self, given_betas=None, beta_schedule="linear", timesteps=1000, + linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3): + super().register_schedule(given_betas, beta_schedule, timesteps, linear_start, linear_end, cosine_s) + + self.shorten_cond_schedule = self.num_timesteps_cond > 1 + if self.shorten_cond_schedule: + self.make_cond_schedule() + + def instantiate_first_stage(self, config): + model = instantiate_from_config(config) + self.first_stage_model = model.eval() + self.first_stage_model.train = disabled_train + for param in self.first_stage_model.parameters(): + param.requires_grad = False + + def instantiate_cond_stage(self, config): + if not self.cond_stage_trainable: + model = instantiate_from_config(config) + self.cond_stage_model = model.eval() + self.cond_stage_model.train = disabled_train + for param in self.cond_stage_model.parameters(): + param.requires_grad = False + else: + model = instantiate_from_config(config) + self.cond_stage_model = model + + def get_learned_conditioning(self, c): + if self.cond_stage_forward is None: + if hasattr(self.cond_stage_model, 'encode') and callable(self.cond_stage_model.encode): + c = self.cond_stage_model.encode(c) + if isinstance(c, DiagonalGaussianDistribution): + c = c.mode() + else: + c = self.cond_stage_model(c) + else: + assert hasattr(self.cond_stage_model, self.cond_stage_forward) + c = getattr(self.cond_stage_model, self.cond_stage_forward)(c) + return c + + def get_first_stage_encoding(self, encoder_posterior, noise=None): + if isinstance(encoder_posterior, DiagonalGaussianDistribution): + z = encoder_posterior.sample(noise=noise) + elif isinstance(encoder_posterior, torch.Tensor): + z = encoder_posterior + else: + raise NotImplementedError(f"encoder_posterior of type '{type(encoder_posterior)}' not yet implemented") + return self.scale_factor * z + + @torch.no_grad() + def encode_first_stage(self, x): + if self.encoder_type == "2d" and x.dim() == 5: + return self.encode_first_stage_2DAE(x) + encoder_posterior = self.first_stage_model.encode(x) + results = self.get_first_stage_encoding(encoder_posterior).detach() + return results + + def encode_first_stage_2DAE(self, x): + """encode frame by frame""" + b, _, t, _, _ = x.shape + results = torch.cat([self.get_first_stage_encoding(self.first_stage_model.encode(x[:,:,i])).detach().unsqueeze(2) for i in range(t)], dim=2) + return results + + def decode_first_stage_2DAE(self, z, **kwargs): + """decode frame by frame""" + _, _, t, _, _ = z.shape + results = torch.cat([self.first_stage_model.decode(z[:,:,i], **kwargs).unsqueeze(2) for i in range(t)], dim=2) + return results + + def _decode_core(self, z, **kwargs): + z = 1. / self.scale_factor * z + + if self.encoder_type == "2d" and z.dim() == 5: + return self.decode_first_stage_2DAE(z) + results = self.first_stage_model.decode(z, **kwargs) + return results + + @torch.no_grad() + def decode_first_stage(self, z, **kwargs): + return self._decode_core(z, **kwargs) + + def differentiable_decode_first_stage(self, z, **kwargs): + """same as decode_first_stage but without decorator""" + return self._decode_core(z, **kwargs) + + @torch.no_grad() + def get_batch_input(self, batch, random_uncond, return_first_stage_outputs=False, return_original_cond=False, is_imgbatch=False): + ## image/video shape: b, c, t, h, w + data_key = 'jpg' if is_imgbatch else self.first_stage_key + x = super().get_input(batch, data_key) + if is_imgbatch: + ## pack image as video + #x = x[:,:,None,:,:] + b = x.shape[0] // self.temporal_length + x = rearrange(x, '(b t) c h w -> b c t h w', b=b, t=self.temporal_length) + x_ori = x + ## encode video frames x to z via a 2D encoder + z = self.encode_first_stage(x) + + ## get caption condition + cond_key = 'txt' if is_imgbatch else self.cond_stage_key + cond = batch[cond_key] + if random_uncond and self.uncond_type == 'empty_seq': + for i, ci in enumerate(cond): + if random.random() < self.uncond_prob: + cond[i] = "" + if isinstance(cond, dict) or isinstance(cond, list): + cond_emb = self.get_learned_conditioning(cond) + else: + cond_emb = self.get_learned_conditioning(cond.to(self.device)) + if random_uncond and self.uncond_type == 'zero_embed': + for i, ci in enumerate(cond): + if random.random() < self.uncond_prob: + cond_emb[i] = torch.zeros_like(ci) + + out = [z, cond_emb] + ## optional output: self-reconst or caption + if return_first_stage_outputs: + xrec = self.decode_first_stage(z) + out.extend([x_ori, xrec]) + if return_original_cond: + out.append(cond) + + return out + + def forward(self, x, c, **kwargs): + if 't' in kwargs: + t = kwargs.pop('t') + else: + t = torch.randint(0, self.num_timesteps, (x.shape[0],), device=self.device).long() + return self.p_losses(x, c, t, **kwargs) + + def shared_step(self, batch, random_uncond, **kwargs): + is_imgbatch = False + if "loader_img" in batch.keys(): + ratio = 10.0 / self.temporal_length + if random.uniform(0.,10.) < ratio: + is_imgbatch = True + batch = batch["loader_img"] + else: + batch = batch["loader_video"] + else: + pass + + x, c = self.get_batch_input(batch, random_uncond=random_uncond, is_imgbatch=is_imgbatch) + loss, loss_dict = self(x, c, is_imgbatch=is_imgbatch, **kwargs) + return loss, loss_dict + + def apply_model(self, x_noisy, t, cond, **kwargs): + if isinstance(cond, dict): + # hybrid case, cond is exptected to be a dict + pass + else: + if not isinstance(cond, list): + cond = [cond] + key = 'c_concat' if self.model.conditioning_key == 'concat' else 'c_crossattn' + cond = {key: cond} + + x_recon = self.model(x_noisy, t, **cond, **kwargs) + + if isinstance(x_recon, tuple): + return x_recon[0] + else: + return x_recon + + def p_losses(self, x_start, cond, t, noise=None, **kwargs): + noise = default(noise, lambda: torch.randn_like(x_start)) + x_noisy = self.q_sample(x_start=x_start, t=t, noise=noise) + if self.frame_cond: + if self.cond_mask.device is not self.device: + self.cond_mask = self.cond_mask.to(self.device) + ## condition on fist few frames + x_noisy = x_start * self.cond_mask + (1.-self.cond_mask) * x_noisy + model_output = self.apply_model(x_noisy, t, cond, **kwargs) + + loss_dict = {} + prefix = 'train' if self.training else 'val' + + if self.parameterization == "x0": + target = x_start + elif self.parameterization == "eps": + target = noise + else: + raise NotImplementedError() + + if self.frame_cond: + ## [b,c,t,h,w]: only care about the predicted part (avoid disturbance) + model_output = model_output[:,:,self.frame_cond:,:,:] + target = target[:,:,self.frame_cond:,:,:] + + loss_simple = self.get_loss(model_output, target, mean=False).mean([1, 2, 3, 4]) + + if torch.isnan(loss_simple).any(): + print(f"loss_simple exists nan: {loss_simple}") + # import pdb; pdb.set_trace() + for i in range(loss_simple.shape[0]): + if torch.isnan(loss_simple[i]).any(): + loss_simple[i] = torch.zeros_like(loss_simple[i]) + + loss_dict.update({f'{prefix}/loss_simple': loss_simple.mean()}) + + if self.logvar.device is not self.device: + self.logvar = self.logvar.to(self.device) + logvar_t = self.logvar[t] + # logvar_t = self.logvar[t.item()].to(self.device) # device conflict when ddp shared + loss = loss_simple / torch.exp(logvar_t) + logvar_t + # loss = loss_simple / torch.exp(self.logvar) + self.logvar + if self.learn_logvar: + loss_dict.update({f'{prefix}/loss_gamma': loss.mean()}) + loss_dict.update({'logvar': self.logvar.data.mean()}) + + loss = self.l_simple_weight * loss.mean() + + if self.original_elbo_weight > 0: + loss_vlb = self.get_loss(model_output, target, mean=False).mean(dim=(1, 2, 3, 4)) + loss_vlb = (self.lvlb_weights[t] * loss_vlb).mean() + loss_dict.update({f'{prefix}/loss_vlb': loss_vlb}) + loss += (self.original_elbo_weight * loss_vlb) + loss_dict.update({f'{prefix}/loss': loss}) + + return loss, loss_dict + + def training_step(self, batch, batch_idx): + loss, loss_dict = self.shared_step(batch, random_uncond=self.classifier_free_guidance) + ## sync_dist | rank_zero_only + self.log_dict(loss_dict, prog_bar=True, logger=True, on_step=True, on_epoch=True, sync_dist=False) + #self.log("epoch/global_step", self.global_step.float(), prog_bar=True, logger=True, on_step=True, on_epoch=False) + ''' + if self.use_scheduler: + lr = self.optimizers().param_groups[0]['lr'] + self.log('lr_abs', lr, prog_bar=True, logger=True, on_step=True, on_epoch=False, rank_zero_only=True) + ''' + if (batch_idx+1) % self.log_every_t == 0: + mainlogger.info(f"batch:{batch_idx}|epoch:{self.current_epoch} [globalstep:{self.global_step}]: loss={loss}") + return loss + + + def _get_denoise_row_from_list(self, samples, desc=''): + denoise_row = [] + for zd in tqdm(samples, desc=desc): + denoise_row.append(self.decode_first_stage(zd.to(self.device))) + n_log_timesteps = len(denoise_row) + + denoise_row = torch.stack(denoise_row) # n_log_timesteps, b, C, H, W + + if denoise_row.dim() == 5: + # img, num_imgs= n_log_timesteps * bs, grid_size=[bs,n_log_timesteps] + # 先batch再n，grid时候一行是一个sample的不同steps，batch是列，行是n + denoise_grid = rearrange(denoise_row, 'n b c h w -> b n c h w') + denoise_grid = rearrange(denoise_grid, 'b n c h w -> (b n) c h w') + denoise_grid = make_grid(denoise_grid, nrow=n_log_timesteps) + elif denoise_row.dim() == 6: + # video, grid_size=[n_log_timesteps*bs, t] + video_length = denoise_row.shape[3] + denoise_grid = rearrange(denoise_row, 'n b c t h w -> b n c t h w') + denoise_grid = rearrange(denoise_grid, 'b n c t h w -> (b n) c t h w') + denoise_grid = rearrange(denoise_grid, 'n c t h w -> (n t) c h w') + denoise_grid = make_grid(denoise_grid, nrow=video_length) + else: + raise ValueError + + return denoise_grid + + @torch.no_grad() + def log_images(self, batch, sample=True, ddim_steps=200, ddim_eta=1., plot_denoise_rows=False, \ + unconditional_guidance_scale=1.0, **kwargs): + """ log images for LatentDiffusion """ + ## TBD: currently, classifier_free_guidance sampling is only supported by DDIM + use_ddim = ddim_steps is not None + log = dict() + z, c, x, xrec, xc = self.get_batch_input(batch, random_uncond=False, + return_first_stage_outputs=True, + return_original_cond=True) + N, _, T, H, W = x.shape + log["inputs"] = x + log["reconst"] = xrec + log["condition"] = xc + + if sample: + # get uncond embedding for classifier-free guidance sampling + if unconditional_guidance_scale != 1.0: + if isinstance(c, dict): + c_cat, c_emb = c["c_concat"][0], c["c_crossattn"][0] + #log["condition_cat"] = c_cat + else: + c_emb = c + + if self.uncond_type == "empty_seq": + prompts = N * [""] + uc = self.get_learned_conditioning(prompts) + elif self.uncond_type == "zero_embed": + uc = torch.zeros_like(c_emb) + ## hybrid case + if isinstance(c, dict): + uc_hybrid = {"c_concat": [c_cat], "c_crossattn": [uc]} + uc = uc_hybrid + else: + uc = None + + with self.ema_scope("Plotting"): + samples, z_denoise_row = self.sample_log(cond=c, batch_size=N, ddim=use_ddim, + ddim_steps=ddim_steps,eta=ddim_eta, + unconditional_guidance_scale=unconditional_guidance_scale, + unconditional_conditioning=uc, mask=self.cond_mask, x0=z, **kwargs) + x_samples = self.decode_first_stage(samples) + log["samples"] = x_samples + + if plot_denoise_rows: + denoise_grid = self._get_denoise_row_from_list(z_denoise_row) + log["denoise_row"] = denoise_grid + + return log + + + def p_mean_variance(self, x, c, t, clip_denoised: bool, return_x0=False, score_corrector=None, corrector_kwargs=None, **kwargs): + t_in = t + model_out = self.apply_model(x, t_in, c, **kwargs) + + if score_corrector is not None: + assert self.parameterization == "eps" + model_out = score_corrector.modify_score(self, model_out, x, t, c, **corrector_kwargs) + + if self.parameterization == "eps": + x_recon = self.predict_start_from_noise(x, t=t, noise=model_out) + elif self.parameterization == "x0": + x_recon = model_out + else: + raise NotImplementedError() + + if clip_denoised: + x_recon.clamp_(-1., 1.) + + model_mean, posterior_variance, posterior_log_variance = self.q_posterior(x_start=x_recon, x_t=x, t=t) + + if return_x0: + return model_mean, posterior_variance, posterior_log_variance, x_recon + else: + return model_mean, posterior_variance, posterior_log_variance + + @torch.no_grad() + def p_sample(self, x, c, t, clip_denoised=False, repeat_noise=False, return_x0=False, \ + temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, **kwargs): + b, *_, device = *x.shape, x.device + outputs = self.p_mean_variance(x=x, c=c, t=t, clip_denoised=clip_denoised, return_x0=return_x0, \ + score_corrector=score_corrector, corrector_kwargs=corrector_kwargs, **kwargs) + if return_x0: + model_mean, _, model_log_variance, x0 = outputs + else: + model_mean, _, model_log_variance = outputs + + noise = noise_like(x.shape, device, repeat_noise) * temperature + if noise_dropout > 0.: + noise = torch.nn.functional.dropout(noise, p=noise_dropout) + # no noise when t == 0 + nonzero_mask = (1 - (t == 0).float()).reshape(b, *((1,) * (len(x.shape) - 1))) + + if return_x0: + return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise, x0 + else: + return model_mean + nonzero_mask * (0.5 * model_log_variance).exp() * noise + + @torch.no_grad() + def p_sample_loop(self, cond, shape, return_intermediates=False, x_T=None, verbose=True, callback=None, \ + timesteps=None, mask=None, x0=None, img_callback=None, start_T=None, log_every_t=None, **kwargs): + + if not log_every_t: + log_every_t = self.log_every_t + device = self.betas.device + b = shape[0] + # sample an initial noise + if x_T is None: + img = torch.randn(shape, device=device) + else: + img = x_T + + intermediates = [img] + if timesteps is None: + timesteps = self.num_timesteps + if start_T is not None: + timesteps = min(timesteps, start_T) + + iterator = tqdm(reversed(range(0, timesteps)), desc='Sampling t', total=timesteps) if verbose else reversed(range(0, timesteps)) + + if mask is not None: + assert x0 is not None + assert x0.shape[2:3] == mask.shape[2:3] # spatial size has to match + + for i in iterator: + ts = torch.full((b,), i, device=device, dtype=torch.long) + if self.shorten_cond_schedule: + assert self.model.conditioning_key != 'hybrid' + tc = self.cond_ids[ts].to(cond.device) + cond = self.q_sample(x_start=cond, t=tc, noise=torch.randn_like(cond)) + + img = self.p_sample(img, cond, ts, clip_denoised=self.clip_denoised, **kwargs) + if mask is not None: + img_orig = self.q_sample(x0, ts) + img = img_orig * mask + (1. - mask) * img + + if i % log_every_t == 0 or i == timesteps - 1: + intermediates.append(img) + if callback: callback(i) + if img_callback: img_callback(img, i) + + if return_intermediates: + return img, intermediates + return img + + @torch.no_grad() + def sample(self, cond, batch_size=16, return_intermediates=False, x_T=None, \ + verbose=True, timesteps=None, mask=None, x0=None, shape=None, **kwargs): + if shape is None: + shape = (batch_size, self.channels, self.temporal_length, *self.image_size) + if cond is not None: + if isinstance(cond, dict): + cond = {key: cond[key][:batch_size] if not isinstance(cond[key], list) else + list(map(lambda x: x[:batch_size], cond[key])) for key in cond} + else: + cond = [c[:batch_size] for c in cond] if isinstance(cond, list) else cond[:batch_size] + return self.p_sample_loop(cond, + shape, + return_intermediates=return_intermediates, x_T=x_T, + verbose=verbose, timesteps=timesteps, + mask=mask, x0=x0, **kwargs) + + @torch.no_grad() + def sample_log(self, cond, batch_size, ddim, ddim_steps, **kwargs): + if ddim: + ddim_sampler = DDIMSampler(self) + shape = (self.channels, self.temporal_length, *self.image_size) + kwargs.update({"clean_cond": True}) + samples, intermediates =ddim_sampler.sample(ddim_steps, batch_size, shape, cond, verbose=False, **kwargs) + + else: + samples, intermediates = self.sample(cond=cond, batch_size=batch_size, return_intermediates=True, **kwargs) + + return samples, intermediates + + def configure_optimizers(self): + """ configure_optimizers for LatentDiffusion """ + lr = self.learning_rate + if self.empty_params_only and hasattr(self, "empty_paras"): + params = [p for n, p in self.model.named_parameters() if n in self.empty_paras] + print('self.empty_paras', len(self.empty_paras)) + for n, p in self.model.named_parameters(): + if n not in self.empty_paras: + p.requires_grad = False + mainlogger.info(f"@Training [{len(params)}] Empty Paramters ONLY.") + else: + params = list(self.model.parameters()) + mainlogger.info(f"@Training [{len(params)}] Full Paramters.") + + if self.learn_logvar: + mainlogger.info('Diffusion model optimizing logvar') + if isinstance(params[0], dict): + params.append({"params": [self.logvar]}) + else: + params.append(self.logvar) + + ## optimizer + optimizer = torch.optim.AdamW(params, lr=lr) + ## lr scheduler + if self.use_scheduler: + mainlogger.info("Setting up LambdaLR scheduler...") + lr_scheduler = self.configure_schedulers(optimizer) + return [optimizer], [lr_scheduler] + + return optimizer + + def configure_schedulers(self, optimizer): + assert 'target' in self.scheduler_config + scheduler_name = self.scheduler_config.target.split('.')[-1] + interval = self.scheduler_config.interval + frequency = self.scheduler_config.frequency + if scheduler_name == "LambdaLRScheduler": + scheduler = instantiate_from_config(self.scheduler_config) + scheduler.start_step = self.global_step + lr_scheduler = { + 'scheduler': LambdaLR(optimizer, lr_lambda=scheduler.schedule), + 'interval': interval, + 'frequency': frequency + } + elif scheduler_name == "CosineAnnealingLRScheduler": + scheduler = instantiate_from_config(self.scheduler_config) + decay_steps = scheduler.decay_steps + last_step = -1 if self.global_step == 0 else scheduler.start_step + lr_scheduler = { + 'scheduler': CosineAnnealingLR(optimizer, T_max=decay_steps, last_epoch=last_step), + 'interval': interval, + 'frequency': frequency + } + else: + raise NotImplementedError + return lr_scheduler + + +class DiffusionWrapper(pl.LightningModule): + def __init__(self, diff_model_config, conditioning_key): + super().__init__() + self.diffusion_model = instantiate_from_config(diff_model_config) + self.conditioning_key = conditioning_key + + def forward(self, x, t, c_concat: list = None, c_crossattn: list = None, + c_adm=None, s=None, mask=None, **kwargs): + # temporal_context = fps is foNone + if self.conditioning_key is None: + out = self.diffusion_model(x, t) + elif self.conditioning_key == 'concat': + xc = torch.cat([x] + c_concat, dim=1) + out = self.diffusion_model(xc, t, **kwargs) + elif self.conditioning_key == 'crossattn': + cc = torch.cat(c_crossattn, 1) + out = self.diffusion_model(x, t, context=cc, **kwargs) + elif self.conditioning_key == 'hybrid': + ## it is just right [b,c,t,h,w]: concatenate in channel dim + xc = torch.cat([x] + c_concat, dim=1) + cc = torch.cat(c_crossattn, 1) + out = self.diffusion_model(xc, t, context=cc, **kwargs) + elif self.conditioning_key == 'resblockcond': + cc = c_crossattn[0] + out = self.diffusion_model(x, t, context=cc) + elif self.conditioning_key == 'adm': + cc = c_crossattn[0] + out = self.diffusion_model(x, t, y=cc) + elif self.conditioning_key == 'hybrid-adm': + assert c_adm is not None + xc = torch.cat([x] + c_concat, dim=1) + cc = torch.cat(c_crossattn, 1) + out = self.diffusion_model(xc, t, context=cc, y=c_adm, **kwargs) + elif self.conditioning_key == 'hybrid-time': + assert s is not None + xc = torch.cat([x] + c_concat, dim=1) + cc = torch.cat(c_crossattn, 1) + out = self.diffusion_model(xc, t, context=cc, s=s) + elif self.conditioning_key == 'concat-time-mask': + # assert s is not None + # mainlogger.info('x & mask:',x.shape,c_concat[0].shape) + xc = torch.cat([x] + c_concat, dim=1) + out = self.diffusion_model(xc, t, context=None, s=s, mask=mask) + elif self.conditioning_key == 'concat-adm-mask': + # assert s is not None + # mainlogger.info('x & mask:',x.shape,c_concat[0].shape) + if c_concat is not None: + xc = torch.cat([x] + c_concat, dim=1) + else: + xc = x + out = self.diffusion_model(xc, t, context=None, y=s, mask=mask) + elif self.conditioning_key == 'hybrid-adm-mask': + cc = torch.cat(c_crossattn, 1) + if c_concat is not None: + xc = torch.cat([x] + c_concat, dim=1) + else: + xc = x + out = self.diffusion_model(xc, t, context=cc, y=s, mask=mask) + elif self.conditioning_key == 'hybrid-time-adm': # adm means y, e.g., class index + # assert s is not None + assert c_adm is not None + xc = torch.cat([x] + c_concat, dim=1) + cc = torch.cat(c_crossattn, 1) + out = self.diffusion_model(xc, t, context=cc, s=s, y=c_adm) + else: + raise NotImplementedError() + + return out \ No newline at end of file diff --git a/lvdm/models/samplers/__init__.py b/lvdm/models/samplers/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/lvdm/models/samplers/ddim.py b/lvdm/models/samplers/ddim.py new file mode 100644 index 0000000000000000000000000000000000000000..b961526e1b7c1f6269b91ef056734e43273bf239 --- /dev/null +++ b/lvdm/models/samplers/ddim.py @@ -0,0 +1,283 @@ +"""SAMPLING ONLY.""" + +import numpy as np +import torch +from tqdm import tqdm + +from lvdm.common import noise_like +from lvdm.models.utils_diffusion import (make_ddim_sampling_parameters, + make_ddim_timesteps) + + +class DDIMSampler(object): + def __init__(self, model, schedule="linear", **kwargs): + super().__init__() + self.model = model + self.ddpm_num_timesteps = model.num_timesteps + self.schedule = schedule + self.counter = 0 + + def register_buffer(self, name, attr): + if type(attr) == torch.Tensor: + if attr.device != torch.device("cuda"): + attr = attr.to(torch.device("cuda")) + setattr(self, name, attr) + + def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True): + self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps, + num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose) + alphas_cumprod = self.model.alphas_cumprod + assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep' + to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device) + + self.register_buffer('betas', to_torch(self.model.betas)) + self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod)) + self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev)) + + # calculations for diffusion q(x_t | x_{t-1}) and others + self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu()))) + self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu()))) + self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu()))) + self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu()))) + self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1))) + + # ddim sampling parameters + ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(), + ddim_timesteps=self.ddim_timesteps, + eta=ddim_eta,verbose=verbose) + self.register_buffer('ddim_sigmas', ddim_sigmas) + self.register_buffer('ddim_alphas', ddim_alphas) + self.register_buffer('ddim_alphas_prev', ddim_alphas_prev) + self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas)) + sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt( + (1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * ( + 1 - self.alphas_cumprod / self.alphas_cumprod_prev)) + self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps) + + @torch.no_grad() + def sample(self, + S, + batch_size, + shape, + conditioning=None, + callback=None, + normals_sequence=None, + img_callback=None, + quantize_x0=False, + eta=0., + mask=None, + x0=None, + temperature=1., + noise_dropout=0., + score_corrector=None, + corrector_kwargs=None, + verbose=True, + schedule_verbose=False, + x_T=None, + log_every_t=100, + unconditional_guidance_scale=1., + unconditional_conditioning=None, + # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ... + **kwargs + ): + + # check condition bs + if conditioning is not None: + if isinstance(conditioning, dict): + try: + cbs = conditioning[list(conditioning.keys())[0]].shape[0] + except: + cbs = conditioning[list(conditioning.keys())[0]][0].shape[0] + + if cbs != batch_size: + print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}") + else: + if conditioning.shape[0] != batch_size: + print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}") + + self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=schedule_verbose) + + # make shape + if len(shape) == 3: + C, H, W = shape + size = (batch_size, C, H, W) + elif len(shape) == 4: + C, T, H, W = shape + size = (batch_size, C, T, H, W) + # print(f'Data shape for DDIM sampling is {size}, eta {eta}') + + samples, intermediates = self.ddim_sampling(conditioning, size, + callback=callback, + img_callback=img_callback, + quantize_denoised=quantize_x0, + mask=mask, x0=x0, + ddim_use_original_steps=False, + noise_dropout=noise_dropout, + temperature=temperature, + score_corrector=score_corrector, + corrector_kwargs=corrector_kwargs, + x_T=x_T, + log_every_t=log_every_t, + unconditional_guidance_scale=unconditional_guidance_scale, + unconditional_conditioning=unconditional_conditioning, + verbose=verbose, + **kwargs) + return samples, intermediates + + @torch.no_grad() + def ddim_sampling(self, cond, shape, + x_T=None, ddim_use_original_steps=False, + callback=None, timesteps=None, quantize_denoised=False, + mask=None, x0=None, img_callback=None, log_every_t=100, + temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, + unconditional_guidance_scale=1., unconditional_conditioning=None, verbose=True, + **kwargs): + device = self.model.betas.device + b = shape[0] + if x_T is None: + img = torch.randn(shape, device=device) + else: + img = x_T + + if timesteps is None: + timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps + elif timesteps is not None and not ddim_use_original_steps: + subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1 + timesteps = self.ddim_timesteps[:subset_end] + + intermediates = {'x_inter': [img], 'pred_x0': [img]} + time_range = reversed(range(0,timesteps)) if ddim_use_original_steps else np.flip(timesteps) + total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0] + if verbose: + iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps) + else: + iterator = time_range + + clean_cond = kwargs.pop("clean_cond", False) + for i, step in enumerate(iterator): + index = total_steps - i - 1 + ts = torch.full((b,), step, device=device, dtype=torch.long) + + # use mask to blend noised original latent (img_orig) & new sampled latent (img) + if mask is not None: + assert x0 is not None + if clean_cond: + img_orig = x0 + else: + img_orig = self.model.q_sample(x0, ts) # TODO: deterministic forward pass? + img = img_orig * mask + (1. - mask) * img # keep original & modify use img + + outs = self.p_sample_ddim(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps, + quantize_denoised=quantize_denoised, temperature=temperature, + noise_dropout=noise_dropout, score_corrector=score_corrector, + corrector_kwargs=corrector_kwargs, + unconditional_guidance_scale=unconditional_guidance_scale, + unconditional_conditioning=unconditional_conditioning, + **kwargs) + + img, pred_x0 = outs + if callback: callback(i) + if img_callback: img_callback(pred_x0, i) + + if index % log_every_t == 0 or index == total_steps - 1: + intermediates['x_inter'].append(img) + intermediates['pred_x0'].append(pred_x0) + + return img, intermediates + + @torch.no_grad() + def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False, + temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None, + unconditional_guidance_scale=1., unconditional_conditioning=None, + uc_type=None, conditional_guidance_scale_temporal=None, **kwargs): + b, *_, device = *x.shape, x.device + if x.dim() == 5: + is_video = True + else: + is_video = False + # f=open('/apdcephfs_cq2/share_1290939/yingqinghe/code/LVDM-private/cfg_range_s5noclamp.txt','a') + # print(f't={t}, model input, min={torch.min(x)}, max={torch.max(x)}',file=f) + if unconditional_conditioning is None or unconditional_guidance_scale == 1.: + e_t = self.model.apply_model(x, t, c, **kwargs) # unet denoiser + else: + # with unconditional condition + if isinstance(c, torch.Tensor): + un_kwargs = kwargs.copy() + if isinstance(unconditional_conditioning, dict): + for uk, uv in unconditional_conditioning.items(): + if uk in un_kwargs: + un_kwargs[uk] = uv + unconditional_conditioning = unconditional_conditioning['uc'] + if 'cond_T' in kwargs and t < kwargs['cond_T']: + if 'features_adapter' in kwargs: + kwargs.pop('features_adapter') + un_kwargs.pop('features_adapter') + # kwargs['features_adapter'] = None + # un_kwargs['features_adapter'] = None + # if 'pose_emb' in kwargs: + # kwargs.pop('pose_emb') + # un_kwargs.pop('pose_emb') + # kwargs['pose_emb'] = None + # un_kwargs['pose_emb'] = None + e_t = self.model.apply_model(x, t, c, **kwargs) + # e_t_uncond = self.model.apply_model(x, t, unconditional_conditioning, **kwargs) + e_t_uncond = self.model.apply_model(x, t, unconditional_conditioning, **un_kwargs) + elif isinstance(c, dict): + e_t = self.model.apply_model(x, t, c, **kwargs) + e_t_uncond = self.model.apply_model(x, t, unconditional_conditioning, **kwargs) + else: + raise NotImplementedError + # text cfg + if uc_type is None: + e_t = e_t_uncond + unconditional_guidance_scale * (e_t - e_t_uncond) + else: + if uc_type == 'cfg_original': + e_t = e_t + unconditional_guidance_scale * (e_t - e_t_uncond) + elif uc_type == 'cfg_ours': + e_t = e_t + unconditional_guidance_scale * (e_t_uncond - e_t) + else: + raise NotImplementedError + # temporal guidance + if conditional_guidance_scale_temporal is not None: + e_t_temporal = self.model.apply_model(x, t, c, **kwargs) + e_t_image = self.model.apply_model(x, t, c, no_temporal_attn=True, **kwargs) + e_t = e_t + conditional_guidance_scale_temporal * (e_t_temporal - e_t_image) + + if score_corrector is not None: + assert self.model.parameterization == "eps" + e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs) + + alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas + alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev + sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas + sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas + # select parameters corresponding to the currently considered timestep + + if is_video: + size = (b, 1, 1, 1, 1) + else: + size = (b, 1, 1, 1) + a_t = torch.full(size, alphas[index], device=device) + a_prev = torch.full(size, alphas_prev[index], device=device) + sigma_t = torch.full(size, sigmas[index], device=device) + sqrt_one_minus_at = torch.full(size, sqrt_one_minus_alphas[index],device=device) + + # current prediction for x_0 + pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt() + # print(f't={t}, pred_x0, min={torch.min(pred_x0)}, max={torch.max(pred_x0)}',file=f) + if quantize_denoised: + pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0) + # direction pointing to x_t + dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t + # # norm pred_x0 + # p=2 + # s=() + # pred_x0 = pred_x0 - torch.max(torch.abs(pred_x0)) + + noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature + if noise_dropout > 0.: + noise = torch.nn.functional.dropout(noise, p=noise_dropout) + + x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise + + return x_prev, pred_x0 diff --git a/lvdm/models/utils_diffusion.py b/lvdm/models/utils_diffusion.py new file mode 100644 index 0000000000000000000000000000000000000000..c57196e60754994ebd6d2afd741607d740604088 --- /dev/null +++ b/lvdm/models/utils_diffusion.py @@ -0,0 +1,105 @@ +import math +import numpy as np +from einops import repeat + +import torch +import torch.nn.functional as F + + +def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False): + """ + Create sinusoidal timestep embeddings. + :param timesteps: a 1-D Tensor of N indices, one per batch element. + These may be fractional. + :param dim: the dimension of the output. + :param max_period: controls the minimum frequency of the embeddings. + :return: an [N x dim] Tensor of positional embeddings. + """ + if not repeat_only: + half = dim // 2 + freqs = torch.exp( + -math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half + ).to(device=timesteps.device) + args = timesteps[:, None].float() * freqs[None] + embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1) + if dim % 2: + embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1) + else: + embedding = repeat(timesteps, 'b -> b d', d=dim) + return embedding + + +def make_beta_schedule(schedule, n_timestep, linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3): + if schedule == "linear": + betas = ( + torch.linspace(linear_start ** 0.5, linear_end ** 0.5, n_timestep, dtype=torch.float64) ** 2 + ) + + elif schedule == "cosine": + timesteps = ( + torch.arange(n_timestep + 1, dtype=torch.float64) / n_timestep + cosine_s + ) + alphas = timesteps / (1 + cosine_s) * np.pi / 2 + alphas = torch.cos(alphas).pow(2) + alphas = alphas / alphas[0] + betas = 1 - alphas[1:] / alphas[:-1] + betas = np.clip(betas, a_min=0, a_max=0.999) + + elif schedule == "sqrt_linear": + betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64) + elif schedule == "sqrt": + betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64) ** 0.5 + else: + raise ValueError(f"schedule '{schedule}' unknown.") + return betas.numpy() + + +def make_ddim_timesteps(ddim_discr_method, num_ddim_timesteps, num_ddpm_timesteps, verbose=True): + if ddim_discr_method == 'uniform': + c = num_ddpm_timesteps // num_ddim_timesteps + ddim_timesteps = np.asarray(list(range(0, num_ddpm_timesteps, c))) + elif ddim_discr_method == 'quad': + ddim_timesteps = ((np.linspace(0, np.sqrt(num_ddpm_timesteps * .8), num_ddim_timesteps)) ** 2).astype(int) + else: + raise NotImplementedError(f'There is no ddim discretization method called "{ddim_discr_method}"') + + # assert ddim_timesteps.shape[0] == num_ddim_timesteps + # add one to get the final alpha values right (the ones from first scale to data during sampling) + steps_out = ddim_timesteps + 1 + if verbose: + print(f'Selected timesteps for ddim sampler: {steps_out}') + return steps_out + + +def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True): + # select alphas for computing the variance schedule + # print(f'ddim_timesteps={ddim_timesteps}, len_alphacums={len(alphacums)}') + alphas = alphacums[ddim_timesteps] + alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist()) + + # according the the formula provided in https://arxiv.org/abs/2010.02502 + sigmas = eta * np.sqrt((1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev)) + if verbose: + print(f'Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}') + print(f'For the chosen value of eta, which is {eta}, ' + f'this results in the following sigma_t schedule for ddim sampler {sigmas}') + return sigmas, alphas, alphas_prev + + +def betas_for_alpha_bar(num_diffusion_timesteps, alpha_bar, max_beta=0.999): + """ + Create a beta schedule that discretizes the given alpha_t_bar function, + which defines the cumulative product of (1-beta) over time from t = [0,1]. + :param num_diffusion_timesteps: the number of betas to produce. + :param alpha_bar: a lambda that takes an argument t from 0 to 1 and + produces the cumulative product of (1-beta) up to that + part of the diffusion process. + :param max_beta: the maximum beta to use; use values lower than 1 to + prevent singularities. + """ + betas = [] + for i in range(num_diffusion_timesteps): + t1 = i / num_diffusion_timesteps + t2 = (i + 1) / num_diffusion_timesteps + betas.append(min(1 - alpha_bar(t2) / alpha_bar(t1), max_beta)) + return np.array(betas) \ No newline at end of file diff --git a/lvdm/modules/attention.py b/lvdm/modules/attention.py new file mode 100644 index 0000000000000000000000000000000000000000..01f593f21f50c3ecc12ad9b328b4ab40ce37422b --- /dev/null +++ b/lvdm/modules/attention.py @@ -0,0 +1,428 @@ +import math +from functools import partial +from inspect import isfunction + +import torch +import torch.nn.functional as F +from einops import rearrange, repeat +from torch import einsum, nn + +try: + import xformers + import xformers.ops + XFORMERS_IS_AVAILBLE = True +except: + XFORMERS_IS_AVAILBLE = False +from lvdm.basics import conv_nd, normalization, zero_module +from lvdm.common import checkpoint, default, exists, init_, max_neg_value, uniq + + +class RelativePosition(nn.Module): + """ https://github.com/evelinehong/Transformer_Relative_Position_PyTorch/blob/master/relative_position.py """ + + def __init__(self, num_units, max_relative_position): + super().__init__() + self.num_units = num_units + self.max_relative_position = max_relative_position + self.embeddings_table = nn.Parameter(torch.Tensor(max_relative_position * 2 + 1, num_units)) + nn.init.xavier_uniform_(self.embeddings_table) + + def forward(self, length_q, length_k): + device = self.embeddings_table.device + range_vec_q = torch.arange(length_q, device=device) + range_vec_k = torch.arange(length_k, device=device) + distance_mat = range_vec_k[None, :] - range_vec_q[:, None] + distance_mat_clipped = torch.clamp(distance_mat, -self.max_relative_position, self.max_relative_position) + final_mat = distance_mat_clipped + self.max_relative_position + # final_mat = th.LongTensor(final_mat).to(self.embeddings_table.device) + # final_mat = th.tensor(final_mat, device=self.embeddings_table.device, dtype=torch.long) + final_mat = final_mat.long() + embeddings = self.embeddings_table[final_mat] + return embeddings + + +class CrossAttention(nn.Module): + + def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0., + relative_position=False, temporal_length=None): + super().__init__() + inner_dim = dim_head * heads + context_dim = default(context_dim, query_dim) + + self.scale = dim_head**-0.5 + self.heads = heads + self.dim_head = dim_head + self.to_q = nn.Linear(query_dim, inner_dim, bias=False) + self.to_k = nn.Linear(context_dim, inner_dim, bias=False) + self.to_v = nn.Linear(context_dim, inner_dim, bias=False) + + self.to_out = nn.Sequential(nn.Linear(inner_dim, query_dim), nn.Dropout(dropout)) + + self.relative_position = relative_position + if self.relative_position: + assert(temporal_length is not None) + self.relative_position_k = RelativePosition(num_units=dim_head, max_relative_position=temporal_length) + self.relative_position_v = RelativePosition(num_units=dim_head, max_relative_position=temporal_length) + else: + ## only used for spatial attention, while NOT for temporal attention + if XFORMERS_IS_AVAILBLE and temporal_length is None: + self.forward = self.efficient_forward + + def forward(self, x, context=None, mask=None): + h = self.heads + + q = self.to_q(x) + context = default(context, x) + k = self.to_k(context) + v = self.to_v(context) + + q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v)) + sim = torch.einsum('b i d, b j d -> b i j', q, k) * self.scale + if self.relative_position: + len_q, len_k, len_v = q.shape[1], k.shape[1], v.shape[1] + k2 = self.relative_position_k(len_q, len_k) + sim2 = einsum('b t d, t s d -> b t s', q, k2) * self.scale # TODO check + sim += sim2 + del q, k + + if exists(mask): + ## feasible for causal attention mask only + max_neg_value = -torch.finfo(sim.dtype).max + mask = repeat(mask, 'b i j -> (b h) i j', h=h) + sim.masked_fill_(~(mask>0.5), max_neg_value) + + # attention, what we cannot get enough of + sim = sim.softmax(dim=-1) + + out = torch.einsum('b i j, b j d -> b i d', sim, v) + if self.relative_position: + v2 = self.relative_position_v(len_q, len_v) + out2 = einsum('b t s, t s d -> b t d', sim, v2) # TODO check + out += out2 + out = rearrange(out, '(b h) n d -> b n (h d)', h=h) + return self.to_out(out) + + def efficient_forward(self, x, context=None, mask=None): + q = self.to_q(x) + context = default(context, x) + k = self.to_k(context) + v = self.to_v(context) + + b, _, _ = q.shape + q, k, v = map( + lambda t: t.unsqueeze(3) + .reshape(b, t.shape[1], self.heads, self.dim_head) + .permute(0, 2, 1, 3) + .reshape(b * self.heads, t.shape[1], self.dim_head) + .contiguous(), + (q, k, v), + ) + # actually compute the attention, what we cannot get enough of + out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None, op=None) + + if exists(mask): + raise NotImplementedError + out = ( + out.unsqueeze(0) + .reshape(b, self.heads, out.shape[1], self.dim_head) + .permute(0, 2, 1, 3) + .reshape(b, out.shape[1], self.heads * self.dim_head) + ) + return self.to_out(out) + + +class BasicTransformerBlock(nn.Module): + + def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=True, + disable_self_attn=False, attention_cls=None): + super().__init__() + attn_cls = CrossAttention if attention_cls is None else attention_cls + self.disable_self_attn = disable_self_attn + self.attn1 = attn_cls(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout, + context_dim=context_dim if self.disable_self_attn else None) + self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff) + self.attn2 = attn_cls(query_dim=dim, context_dim=context_dim, heads=n_heads, dim_head=d_head, dropout=dropout) + self.norm1 = nn.LayerNorm(dim) + self.norm2 = nn.LayerNorm(dim) + self.norm3 = nn.LayerNorm(dim) + self.checkpoint = checkpoint + + def forward(self, x, context=None, mask=None): + ## implementation tricks: because checkpointing doesn't support non-tensor (e.g. None or scalar) arguments + input_tuple = (x,) ## should not be (x), otherwise *input_tuple will decouple x into multiple arguments + if context is not None: + input_tuple = (x, context) + if mask is not None: + forward_mask = partial(self._forward, mask=mask) + return checkpoint(forward_mask, (x,), self.parameters(), self.checkpoint) + + # It seems that it will not be executed forever + if context is not None and mask is not None: + input_tuple = (x, context, mask) + return checkpoint(self._forward, input_tuple, self.parameters(), self.checkpoint) + + def _forward(self, x, context=None, mask=None): + x = self.attn1(self.norm1(x), context=context if self.disable_self_attn else None, mask=mask) + x + x = self.attn2(self.norm2(x), context=context, mask=mask) + x + x = self.ff(self.norm3(x)) + x + return x + + +class SpatialTransformer(nn.Module): + """ + Transformer block for image-like data in spatial axis. + First, project the input (aka embedding) + and reshape to b, t, d. + Then apply standard transformer action. + Finally, reshape to image + NEW: use_linear for more efficiency instead of the 1x1 convs + """ + + def __init__(self, in_channels, n_heads, d_head, depth=1, dropout=0., context_dim=None, + use_checkpoint=True, disable_self_attn=False, use_linear=False): + super().__init__() + self.in_channels = in_channels + inner_dim = n_heads * d_head + self.norm = torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True) + if not use_linear: + self.proj_in = nn.Conv2d(in_channels, inner_dim, kernel_size=1, stride=1, padding=0) + else: + self.proj_in = nn.Linear(in_channels, inner_dim) + + self.transformer_blocks = nn.ModuleList([ + BasicTransformerBlock( + inner_dim, + n_heads, + d_head, + dropout=dropout, + context_dim=context_dim, + disable_self_attn=disable_self_attn, + checkpoint=use_checkpoint) for d in range(depth) + ]) + if not use_linear: + self.proj_out = zero_module(nn.Conv2d(inner_dim, in_channels, kernel_size=1, stride=1, padding=0)) + else: + self.proj_out = zero_module(nn.Linear(inner_dim, in_channels)) + self.use_linear = use_linear + + + def forward(self, x, context=None): + b, c, h, w = x.shape + x_in = x + x = self.norm(x) + if not self.use_linear: + x = self.proj_in(x) + x = rearrange(x, 'b c h w -> b (h w) c').contiguous() + if self.use_linear: + x = self.proj_in(x) + for i, block in enumerate(self.transformer_blocks): + x = block(x, context=context) + if self.use_linear: + x = self.proj_out(x) + x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w).contiguous() + if not self.use_linear: + x = self.proj_out(x) + return x + x_in + + +class TemporalTransformer(nn.Module): + """ + Transformer block for image-like data in temporal axis. + First, reshape to b, t, d. + Then apply standard transformer action. + Finally, reshape to image + """ + def __init__(self, in_channels, n_heads, d_head, depth=1, dropout=0., context_dim=None, + use_checkpoint=True, use_linear=False, only_self_att=True, causal_attention=False, + relative_position=False, temporal_length=None, use_image_dataset=False): + super().__init__() + self.only_self_att = only_self_att + self.relative_position = relative_position + self.causal_attention = causal_attention + self.use_image_dataset = use_image_dataset + self.in_channels = in_channels + inner_dim = n_heads * d_head + self.norm = torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True) + self.proj_in = nn.Conv1d(in_channels, inner_dim, kernel_size=1, stride=1, padding=0) + if not use_linear: + self.proj_in = nn.Conv1d(in_channels, inner_dim, kernel_size=1, stride=1, padding=0) + else: + self.proj_in = nn.Linear(in_channels, inner_dim) + + if relative_position: + assert(temporal_length is not None) + attention_cls = partial(CrossAttention, relative_position=True, temporal_length=temporal_length) + else: + attention_cls = None + + if self.only_self_att: + context_dim = None + self.transformer_blocks = nn.ModuleList([ + BasicTransformerBlock( + inner_dim, + n_heads, + d_head, + dropout=dropout, + context_dim=context_dim, + attention_cls=attention_cls, + checkpoint=use_checkpoint) for d in range(depth) + ]) + if not use_linear: + self.proj_out = zero_module(nn.Conv1d(inner_dim, in_channels, kernel_size=1, stride=1, padding=0)) + else: + self.proj_out = zero_module(nn.Linear(inner_dim, in_channels)) + self.use_linear = use_linear + + def forward(self, x, context=None, is_imgbatch=False): + b, c, t, h, w = x.shape + x_in = x + x = self.norm(x) + x = rearrange(x, 'b c t h w -> (b h w) c t').contiguous() + if not self.use_linear: + x = self.proj_in(x) + x = rearrange(x, 'bhw c t -> bhw t c').contiguous() + if self.use_linear: + x = self.proj_in(x) + + temp_mask = None + if self.causal_attention: + temp_mask = torch.tril(torch.ones([1, t, t])) + if is_imgbatch: + temp_mask = torch.eye(t).unsqueeze(0) + if temp_mask is not None: + mask = temp_mask.to(x.device) + mask = repeat(mask, 'l i j -> (l bhw) i j', bhw=b*h*w) + else: + mask = None + + if self.only_self_att: + ## note: if no context is given, cross-attention defaults to self-attention + for i, block in enumerate(self.transformer_blocks): + x = block(x, mask=mask) + x = rearrange(x, '(b hw) t c -> b hw t c', b=b).contiguous() + else: + x = rearrange(x, '(b hw) t c -> b hw t c', b=b).contiguous() + context = rearrange(context, '(b t) l con -> b t l con', t=t).contiguous() + for i, block in enumerate(self.transformer_blocks): + # calculate each batch one by one (since number in shape could not greater then 65,535 for some package) + for j in range(b): + unit_context = context[j][0:1] + context_j = repeat(unit_context, 't l con -> (t r) l con', r=(h * w)).contiguous() + ## note: causal mask will not applied in cross-attention case + x[j] = block(x[j], context=context_j) + + if self.use_linear: + x = self.proj_out(x) + x = rearrange(x, 'b (h w) t c -> b c t h w', h=h, w=w).contiguous() + if not self.use_linear: + x = rearrange(x, 'b hw t c -> (b hw) c t').contiguous() + x = self.proj_out(x) + x = rearrange(x, '(b h w) c t -> b c t h w', b=b, h=h, w=w).contiguous() + + if self.use_image_dataset: + x = 0.0 * x + x_in + else: + x = x + x_in + return x + + +class GEGLU(nn.Module): + def __init__(self, dim_in, dim_out): + super().__init__() + self.proj = nn.Linear(dim_in, dim_out * 2) + + def forward(self, x): + x, gate = self.proj(x).chunk(2, dim=-1) + return x * F.gelu(gate) + + +class FeedForward(nn.Module): + def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.): + super().__init__() + inner_dim = int(dim * mult) + dim_out = default(dim_out, dim) + project_in = nn.Sequential( + nn.Linear(dim, inner_dim), + nn.GELU() + ) if not glu else GEGLU(dim, inner_dim) + + self.net = nn.Sequential( + project_in, + nn.Dropout(dropout), + nn.Linear(inner_dim, dim_out) + ) + + def forward(self, x): + return self.net(x) + + +class LinearAttention(nn.Module): + def __init__(self, dim, heads=4, dim_head=32): + super().__init__() + self.heads = heads + hidden_dim = dim_head * heads + self.to_qkv = nn.Conv2d(dim, hidden_dim * 3, 1, bias = False) + self.to_out = nn.Conv2d(hidden_dim, dim, 1) + + def forward(self, x): + b, c, h, w = x.shape + qkv = self.to_qkv(x) + q, k, v = rearrange(qkv, 'b (qkv heads c) h w -> qkv b heads c (h w)', heads = self.heads, qkv=3) + k = k.softmax(dim=-1) + context = torch.einsum('bhdn,bhen->bhde', k, v) + out = torch.einsum('bhde,bhdn->bhen', context, q) + out = rearrange(out, 'b heads c (h w) -> b (heads c) h w', heads=self.heads, h=h, w=w) + return self.to_out(out) + + +class SpatialSelfAttention(nn.Module): + def __init__(self, in_channels): + super().__init__() + self.in_channels = in_channels + + self.norm = torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True) + self.q = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0) + self.k = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0) + self.v = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0) + self.proj_out = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0) + + def forward(self, x): + h_ = x + h_ = self.norm(h_) + q = self.q(h_) + k = self.k(h_) + v = self.v(h_) + + # compute attention + b,c,h,w = q.shape + q = rearrange(q, 'b c h w -> b (h w) c') + k = rearrange(k, 'b c h w -> b c (h w)') + w_ = torch.einsum('bij,bjk->bik', q, k) + + w_ = w_ * (int(c)**(-0.5)) + w_ = torch.nn.functional.softmax(w_, dim=2) + + # attend to values + v = rearrange(v, 'b c h w -> b c (h w)') + w_ = rearrange(w_, 'b i j -> b j i') + h_ = torch.einsum('bij,bjk->bik', v, w_) + h_ = rearrange(h_, 'b c (h w) -> b c h w', h=h) + h_ = self.proj_out(h_) + + return x+h_ diff --git a/lvdm/modules/attention_temporal.py b/lvdm/modules/attention_temporal.py new file mode 100644 index 0000000000000000000000000000000000000000..7cee2e83f8750d6b3a217c350199657e9de2fa1b --- /dev/null +++ b/lvdm/modules/attention_temporal.py @@ -0,0 +1,995 @@ +import math +from inspect import isfunction + +import torch +import torch as th +from torch import nn, einsum +import torch.nn.functional as F +from einops import rearrange, repeat +try: + import xformers + import xformers.ops + XFORMERS_IS_AVAILBLE = True +except: + XFORMERS_IS_AVAILBLE = False + +from lvdm.common import ( + checkpoint, + exists, + uniq, + default, + max_neg_value, + init_ +) +from lvdm.basics import ( + conv_nd, + zero_module, + normalization +) + + +class GEGLU(nn.Module): + def __init__(self, dim_in, dim_out): + super().__init__() + self.proj = nn.Linear(dim_in, dim_out * 2) + + def forward(self, x): + x, gate = self.proj(x).chunk(2, dim=-1) + return x * F.gelu(gate) + + +class FeedForward(nn.Module): + def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.): + super().__init__() + inner_dim = int(dim * mult) + dim_out = default(dim_out, dim) + project_in = nn.Sequential( + nn.Linear(dim, inner_dim), + nn.GELU() + ) if not glu else GEGLU(dim, inner_dim) + + self.net = nn.Sequential( + project_in, + nn.Dropout(dropout), + nn.Linear(inner_dim, dim_out) + ) + + def forward(self, x): + return self.net(x) + + +def Normalize(in_channels): + return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True) + + +# --------------------------------------------------------------------------------------------------- +class RelativePosition(nn.Module): + """ https://github.com/evelinehong/Transformer_Relative_Position_PyTorch/blob/master/relative_position.py """ + + def __init__(self, num_units, max_relative_position): + super().__init__() + self.num_units = num_units + self.max_relative_position = max_relative_position + self.embeddings_table = nn.Parameter(th.Tensor(max_relative_position * 2 + 1, num_units)) + nn.init.xavier_uniform_(self.embeddings_table) + + def forward(self, length_q, length_k): + device = self.embeddings_table.device + range_vec_q = th.arange(length_q, device=device) + range_vec_k = th.arange(length_k, device=device) + distance_mat = range_vec_k[None, :] - range_vec_q[:, None] + distance_mat_clipped = th.clamp(distance_mat, -self.max_relative_position, self.max_relative_position) + final_mat = distance_mat_clipped + self.max_relative_position + # final_mat = th.LongTensor(final_mat).to(self.embeddings_table.device) + # final_mat = th.tensor(final_mat, device=self.embeddings_table.device, dtype=torch.long) + final_mat = final_mat.long() + embeddings = self.embeddings_table[final_mat] + return embeddings + + +class TemporalCrossAttention(nn.Module): + def __init__(self, + query_dim, + context_dim=None, + heads=8, + dim_head=64, + dropout=0., + temporal_length=None, # For relative positional representation and image-video joint training. + image_length=None, # For image-video joint training. + use_relative_position=False, # whether use relative positional representation in temporal attention. + img_video_joint_train=False, # For image-video joint training. + use_tempoal_causal_attn=False, + bidirectional_causal_attn=False, + tempoal_attn_type=None, + joint_train_mode="same_batch", + **kwargs, + ): + super().__init__() + inner_dim = dim_head * heads + context_dim = default(context_dim, query_dim) + self.context_dim = context_dim + + self.scale = dim_head ** -0.5 + self.heads = heads + self.temporal_length = temporal_length + self.use_relative_position = use_relative_position + self.img_video_joint_train = img_video_joint_train + self.bidirectional_causal_attn = bidirectional_causal_attn + self.joint_train_mode = joint_train_mode + assert(joint_train_mode in ["same_batch", "diff_batch"]) + self.tempoal_attn_type = tempoal_attn_type + + if bidirectional_causal_attn: + assert use_tempoal_causal_attn + if tempoal_attn_type: + assert(tempoal_attn_type in ['sparse_causal', 'sparse_causal_first']) + assert(not use_tempoal_causal_attn) + assert(not (img_video_joint_train and (self.joint_train_mode == "same_batch"))) + self.to_q = nn.Linear(query_dim, inner_dim, bias=False) + self.to_k = nn.Linear(context_dim, inner_dim, bias=False) + self.to_v = nn.Linear(context_dim, inner_dim, bias=False) + + assert(not (img_video_joint_train and (self.joint_train_mode == "same_batch") and use_tempoal_causal_attn)) + if img_video_joint_train: + if self.joint_train_mode == "same_batch": + mask = torch.ones([1, temporal_length+image_length, temporal_length+image_length]) + # mask[:, image_length:, :] = 0 + # mask[:, :, image_length:] = 0 + mask[:, temporal_length:, :] = 0 + mask[:, :, temporal_length:] = 0 + self.mask = mask + else: + self.mask = None + elif use_tempoal_causal_attn: + # normal causal attn + self.mask = torch.tril(torch.ones([1, temporal_length, temporal_length])) + elif tempoal_attn_type == 'sparse_causal': + # all frames interact with only the `prev` & self frame + mask1 = torch.tril(torch.ones([1, temporal_length, temporal_length])).bool() # true indicates keeping + mask2 = torch.zeros([1, temporal_length, temporal_length]) # initialize to same shape with mask1 + mask2[:,2:temporal_length, :temporal_length-2] = torch.tril(torch.ones([1,temporal_length-2, temporal_length-2])) + mask2=(1-mask2).bool() # false indicates masking + self.mask = mask1 & mask2 + elif tempoal_attn_type == 'sparse_causal_first': + # all frames interact with only the `first` & self frame + mask1 = torch.tril(torch.ones([1, temporal_length, temporal_length])).bool() # true indicates keeping + mask2 = torch.zeros([1, temporal_length, temporal_length]) + mask2[:,2:temporal_length, 1:temporal_length-1] = torch.tril(torch.ones([1,temporal_length-2, temporal_length-2])) + mask2=(1-mask2).bool() # false indicates masking + self.mask = mask1 & mask2 + else: + self.mask = None + + if use_relative_position: + assert(temporal_length is not None) + self.relative_position_k = RelativePosition(num_units=dim_head, max_relative_position=temporal_length) + self.relative_position_v = RelativePosition(num_units=dim_head, max_relative_position=temporal_length) + + self.to_out = nn.Sequential( + nn.Linear(inner_dim, query_dim), + nn.Dropout(dropout) + ) + + nn.init.constant_(self.to_q.weight, 0) + nn.init.constant_(self.to_k.weight, 0) + nn.init.constant_(self.to_v.weight, 0) + nn.init.constant_(self.to_out[0].weight, 0) + nn.init.constant_(self.to_out[0].bias, 0) + + def forward(self, x, context=None, mask=None): + # if context is None: + # print(f'[Temp Attn] x={x.shape},context=None') + # else: + # print(f'[Temp Attn] x={x.shape},context={context.shape}') + + nh = self.heads + out = x + q = self.to_q(out) + # if context is not None: + # print(f'temporal context 1 ={context.shape}') + # print(f'x={x.shape}') + context = default(context, x) + # print(f'temporal context 2 ={context.shape}') + k = self.to_k(context) + v = self.to_v(context) + # print(f'q ={q.shape},k={k.shape}') + + q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=nh), (q, k, v)) + sim = einsum('b i d, b j d -> b i j', q, k) * self.scale + + if self.use_relative_position: + len_q, len_k, len_v = q.shape[1], k.shape[1], v.shape[1] + k2 = self.relative_position_k(len_q, len_k) + sim2 = einsum('b t d, t s d -> b t s', q, k2) * self.scale # TODO check + sim += sim2 + # print('mask',mask) + if exists(self.mask): + if mask is None: + mask = self.mask.to(sim.device) + else: + mask = self.mask.to(sim.device).bool() & mask #.to(sim.device) + else: + mask = mask + # if self.img_video_joint_train: + # # process mask (make mask same shape with sim) + # c, h, w = mask.shape + # c, t, s = sim.shape + # # assert(h == w and t == s),f"mask={mask.shape}, sim={sim.shape}, h={h}, w={w}, t={t}, s={s}" + + # if h > t: + # mask = mask[:, :t, :] + # elif h < t: # pad zeros to mask (no attention) only initial mask =1 area compute weights + # mask_ = torch.zeros([c,t,w]).to(mask.device) + # mask_[:, :h, :] = mask + # mask = mask_ + # c, h, w = mask.shape + # if w > s: + # mask = mask[:, :, :s] + # elif w < s: # pad zeros to mask + # mask_ = torch.zeros([c,h,s]).to(mask.device) + # mask_[:, :, :w] = mask + # mask = mask_ + + # max_neg_value = -torch.finfo(sim.dtype).max + # sim = sim.float().masked_fill(mask == 0, max_neg_value) + if mask is not None: + max_neg_value = -1e9 + sim = sim + (1-mask.float()) * max_neg_value # 1=masking,0=no masking + # print('sim after masking: ', sim) + + # if torch.isnan(sim).any() or torch.isinf(sim).any() or (not sim.any()): + # print(f'sim [after masking], isnan={torch.isnan(sim).any()}, isinf={torch.isinf(sim).any()}, allzero={not sim.any()}') + + attn = sim.softmax(dim=-1) + # print('attn after softmax: ', attn) + # if torch.isnan(attn).any() or torch.isinf(attn).any() or (not attn.any()): + # print(f'attn [after softmax], isnan={torch.isnan(attn).any()}, isinf={torch.isinf(attn).any()}, allzero={not attn.any()}') + + # attn = torch.where(torch.isnan(attn), torch.full_like(attn,0), attn) + # if torch.isinf(attn.detach()).any(): + # import pdb;pdb.set_trace() + # if torch.isnan(attn.detach()).any(): + # import pdb;pdb.set_trace() + out = einsum('b i j, b j d -> b i d', attn, v) + + if self.bidirectional_causal_attn: + mask_reverse = torch.triu(torch.ones([1, self.temporal_length, self.temporal_length], device=sim.device)) + sim_reverse = sim.float().masked_fill(mask_reverse == 0, max_neg_value) + attn_reverse = sim_reverse.softmax(dim=-1) + out_reverse = einsum('b i j, b j d -> b i d', attn_reverse, v) + out += out_reverse + + if self.use_relative_position: + v2 = self.relative_position_v(len_q, len_v) + out2 = einsum('b t s, t s d -> b t d', attn, v2) # TODO check + out += out2 # TODO check：先add还是先merge head？先计算rpr，on split head之后的数据，然后再merge。 + out = rearrange(out, '(b h) n d -> b n (h d)', h=nh) # merge head + return self.to_out(out) + + +class CrossAttention(nn.Module): + def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0., + sa_shared_kv=False, shared_type='only_first', **kwargs,): + super().__init__() + inner_dim = dim_head * heads + context_dim = default(context_dim, query_dim) + self.sa_shared_kv = sa_shared_kv + assert(shared_type in ['only_first', 'all_frames', 'first_and_prev', 'only_prev', 'full', 'causal', 'full_qkv']) + self.shared_type = shared_type + + self.dim_head = dim_head + self.scale = dim_head ** -0.5 + self.heads = heads + + self.to_q = nn.Linear(query_dim, inner_dim, bias=False) + self.to_k = nn.Linear(context_dim, inner_dim, bias=False) + self.to_v = nn.Linear(context_dim, inner_dim, bias=False) + + self.to_out = nn.Sequential( + nn.Linear(inner_dim, query_dim), + nn.Dropout(dropout) + ) + if XFORMERS_IS_AVAILBLE: + self.forward = self.efficient_forward + + def forward(self, x, context=None, mask=None): + h = self.heads + b = x.shape[0] + + q = self.to_q(x) + context = default(context, x) + k = self.to_k(context) + v = self.to_v(context) + if self.sa_shared_kv: + if self.shared_type == 'only_first': + k,v = map(lambda xx: rearrange(xx[0].unsqueeze(0), 'b n c -> (b n) c').unsqueeze(0).repeat(b,1,1), + (k,v)) + else: + raise NotImplementedError + + q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v)) + + sim = einsum('b i d, b j d -> b i j', q, k) * self.scale + + if exists(mask): + mask = rearrange(mask, 'b ... -> b (...)') + max_neg_value = -torch.finfo(sim.dtype).max + mask = repeat(mask, 'b j -> (b h) () j', h=h) + sim.masked_fill_(~mask, max_neg_value) + + # attention, what we cannot get enough of + attn = sim.softmax(dim=-1) + + out = einsum('b i j, b j d -> b i d', attn, v) + out = rearrange(out, '(b h) n d -> b n (h d)', h=h) + return self.to_out(out) + + def efficient_forward(self, x, context=None, mask=None): + q = self.to_q(x) + context = default(context, x) + k = self.to_k(context) + v = self.to_v(context) + + b, _, _ = q.shape + q, k, v = map( + lambda t: t.unsqueeze(3) + .reshape(b, t.shape[1], self.heads, self.dim_head) + .permute(0, 2, 1, 3) + .reshape(b * self.heads, t.shape[1], self.dim_head) + .contiguous(), + (q, k, v), + ) + # actually compute the attention, what we cannot get enough of + out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None, op=None) + + if exists(mask): + raise NotImplementedError + out = ( + out.unsqueeze(0) + .reshape(b, self.heads, out.shape[1], self.dim_head) + .permute(0, 2, 1, 3) + .reshape(b, out.shape[1], self.heads * self.dim_head) + ) + return self.to_out(out) + +class VideoSpatialCrossAttention(CrossAttention): + def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0): + super().__init__(query_dim, context_dim, heads, dim_head, dropout) + def forward(self, x, context=None, mask=None): + b, c, t, h, w = x.shape + if context is not None: + context = context.repeat(t, 1, 1) + x = super.forward(spatial_attn_reshape(x), context=context) + x + return spatial_attn_reshape_back(x,b,h) + +class BasicTransformerBlockST(nn.Module): + def __init__(self, + # Spatial Stuff + dim, + n_heads, + d_head, + dropout=0., + context_dim=None, + gated_ff=True, + checkpoint=True, + # Temporal Stuff + temporal_length=None, + image_length=None, + use_relative_position=True, + img_video_joint_train=False, + cross_attn_on_tempoal=False, + temporal_crossattn_type="selfattn", + order="stst", + temporalcrossfirst=False, + temporal_context_dim=None, + split_stcontext=False, + local_spatial_temporal_attn=False, + window_size=2, + **kwargs, + ): + super().__init__() + # Self attention + self.attn1 = CrossAttention(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout, **kwargs,) + self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff) + # cross attention if context is not None + self.attn2 = CrossAttention(query_dim=dim, context_dim=context_dim, + heads=n_heads, dim_head=d_head, dropout=dropout, **kwargs,) + self.norm1 = nn.LayerNorm(dim) + self.norm2 = nn.LayerNorm(dim) + self.norm3 = nn.LayerNorm(dim) + self.checkpoint = checkpoint + self.order = order + assert(self.order in ["stst", "sstt", "st_parallel"]) + self.temporalcrossfirst = temporalcrossfirst + self.split_stcontext = split_stcontext + self.local_spatial_temporal_attn = local_spatial_temporal_attn + if self.local_spatial_temporal_attn: + assert(self.order == 'stst') + assert(self.order == 'stst') + self.window_size = window_size + if not split_stcontext: + temporal_context_dim = context_dim + # Temporal attention + assert(temporal_crossattn_type in ["selfattn", "crossattn", "skip"]) + self.temporal_crossattn_type = temporal_crossattn_type + self.attn1_tmp = TemporalCrossAttention(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout, + temporal_length=temporal_length, + image_length=image_length, + use_relative_position=use_relative_position, + img_video_joint_train=img_video_joint_train, + **kwargs, + ) + self.attn2_tmp = TemporalCrossAttention(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout, + # cross attn + context_dim=temporal_context_dim if temporal_crossattn_type == "crossattn" else None, + # temporal attn + temporal_length=temporal_length, + image_length=image_length, + use_relative_position=use_relative_position, + img_video_joint_train=img_video_joint_train, + **kwargs, + ) + self.norm4 = nn.LayerNorm(dim) + self.norm5 = nn.LayerNorm(dim) + # self.norm1_tmp = nn.LayerNorm(dim) + # self.norm2_tmp = nn.LayerNorm(dim) + + ############################################################################################################################################## + def forward(self, x, context=None, temporal_context=None, no_temporal_attn=None, attn_mask=None, **kwargs): + # print(f'no_temporal_attn={no_temporal_attn}') + + if not self.split_stcontext: + # st cross attention use the same context vector + temporal_context = context.detach().clone() + + if context is None and temporal_context is None: + # self-attention models + if no_temporal_attn: + raise NotImplementedError + return checkpoint(self._forward_nocontext, (x), self.parameters(), self.checkpoint) + else: + # cross-attention models + if no_temporal_attn: + forward_func = self._forward_no_temporal_attn + else: + forward_func = self._forward + inputs = (x, context, temporal_context) if temporal_context is not None else (x, context) + return checkpoint(forward_func, inputs, self.parameters(), self.checkpoint) + # if attn_mask is not None: + # return checkpoint(self._forward, (x, context, temporal_context, attn_mask), self.parameters(), self.checkpoint) + # return checkpoint(self._forward, (x, context, temporal_context), self.parameters(), self.checkpoint) + + def _forward(self, x, context=None, temporal_context=None, mask=None, no_temporal_attn=None, ): + assert(x.dim() == 5), f"x shape = {x.shape}" + b, c, t, h, w = x.shape + + if self.order in ["stst", "sstt"]: + x = self._st_cross_attn(x, context, temporal_context=temporal_context, order=self.order, mask=mask,)#no_temporal_attn=no_temporal_attn, + elif self.order == "st_parallel": + x = self._st_cross_attn_parallel(x, context, temporal_context=temporal_context, order=self.order,)#no_temporal_attn=no_temporal_attn, + else: + raise NotImplementedError + + x = self.ff(self.norm3(x)) + x + if (no_temporal_attn is None) or (not no_temporal_attn): + x = rearrange(x, '(b h w) t c -> b c t h w', b=b,h=h,w=w) # 3d -> 5d + elif no_temporal_attn: + x = rearrange(x, '(b t) (h w) c -> b c t h w', b=b,h=h,w=w) # 3d -> 5d + return x + + def _forward_no_temporal_attn(self, x, context=None, temporal_context=None, ): + # temporary implementation :( + # because checkpoint does not support non-tensor inputs currently. + assert(x.dim() == 5), f"x shape = {x.shape}" + b, c, t, h, w = x.shape + + if self.order in ["stst", "sstt"]: + # x = self._st_cross_attn(x, context, temporal_context=temporal_context, order=self.order, no_temporal_attn=True,) + # mask = torch.zeros([1, t, t], device=x.device).bool() if context is None else torch.zeros([1, context.shape[1], t], device=x.device).bool() + mask = torch.zeros([1, t, t], device=x.device).bool() + x = self._st_cross_attn(x, context, temporal_context=temporal_context, order=self.order, mask=mask,) + elif self.order == "st_parallel": + x = self._st_cross_attn_parallel(x, context, temporal_context=temporal_context, order=self.order, no_temporal_attn=True,) + else: + raise NotImplementedError + + x = self.ff(self.norm3(x)) + x + x = rearrange(x, '(b h w) t c -> b c t h w', b=b,h=h,w=w) # 3d -> 5d + # x = rearrange(x, '(b t) (h w) c -> b c t h w', b=b,h=h,w=w) # 3d -> 5d + return x + + def _forward_nocontext(self, x, no_temporal_attn=None): + assert(x.dim() == 5), f"x shape = {x.shape}" + b, c, t, h, w = x.shape + + if self.order in ["stst", "sstt"]: + x = self._st_cross_attn(x, order=self.order, no_temporal_attn=no_temporal_attn) + elif self.order == "st_parallel": + x = self._st_cross_attn_parallel(x, order=self.order, no_temporal_attn=no_temporal_attn) + else: + raise NotImplementedError + + x = self.ff(self.norm3(x)) + x + x = rearrange(x, '(b h w) t c -> b c t h w', b=b,h=h,w=w) # 3d -> 5d + + return x + ############################################################################################################################################## + + def _st_cross_attn(self, x, context=None, temporal_context=None, order="stst", mask=None): #no_temporal_attn=None, + b, c, t, h, w = x.shape + # print(f'[_st_cross_attn input] x={x.shape}, context={context.shape}') + + if order == "stst": + # spatial self attention + x = rearrange(x, 'b c t h w -> (b t) (h w) c') + x = self.attn1(self.norm1(x)) + x + x = rearrange(x, '(b t) (h w) c -> b c t h w', b=b,h=h) + + # temporal self attention + # if (no_temporal_attn is None) or (not no_temporal_attn): + if self.local_spatial_temporal_attn: + x = local_spatial_temporal_attn_reshape(x,window_size=self.window_size) + else: + x = rearrange(x, 'b c t h w -> (b h w) t c') + x = self.attn1_tmp(self.norm4(x), mask=mask) + x + + if self.local_spatial_temporal_attn: + x = local_spatial_temporal_attn_reshape_back(x, window_size=self.window_size, + b=b, h=h, w=w, t=t) + else: + x = rearrange(x, '(b h w) t c -> b c t h w', b=b,h=h,w=w) # 3d -> 5d + + # spatial cross attention + x = rearrange(x, 'b c t h w -> (b t) (h w) c') + # context_ = context.repeat(t, 1, 1) if context is not None else None + # print(f'[before spatial cross] context={context.shape}') + if context is not None: + if context.shape[0] == t: # img captions no_temporal_attn or + context_ = context + else: + context_ = [] + for i in range(context.shape[0]): + context_.append(context[i].unsqueeze(0).repeat(t, 1, 1)) + context_ = torch.cat(context_,dim=0) + else: + context_ = None + # print(f'[before spatial cross] x={x.shape}, context_={context_.shape}') + x = self.attn2(self.norm2(x), context=context_) + x + + # temporal cross attention + # if (no_temporal_attn is None) or (not no_temporal_attn): + x = rearrange(x, '(b t) (h w) c -> b c t h w', b=b,h=h) + x = rearrange(x, 'b c t h w -> (b h w) t c') + if self.temporal_crossattn_type == "crossattn": + # tmporal cross attention + if temporal_context is not None: + # print(f'STATTN context={context.shape}, temporal_context={temporal_context.shape}') + temporal_context = torch.cat([context, temporal_context], dim=1) # blc + # print(f'STATTN after concat temporal_context={temporal_context.shape}') + temporal_context = temporal_context.repeat(h*w, 1,1) + # print(f'after repeat temporal_context={temporal_context.shape}') + else: + temporal_context = context[0:1,...].repeat(h*w, 1, 1) + # print(f'STATTN after concat x={x.shape}') + x = self.attn2_tmp(self.norm5(x), context=temporal_context, mask=mask) + x + elif self.temporal_crossattn_type == "selfattn": + # temporal self attention + x = self.attn2_tmp(self.norm5(x), context=None, mask=mask) + x + elif self.temporal_crossattn_type == "skip": + # no temporal cross and self attention + pass + else: + raise NotImplementedError + + elif order == "sstt": + # spatial self attention + x = rearrange(x, 'b c t h w -> (b t) (h w) c') + x = self.attn1(self.norm1(x)) + x + + # spatial cross attention + context_ = context.repeat(t, 1, 1) if context is not None else None + x = self.attn2(self.norm2(x), context=context_) + x + x = rearrange(x, '(b t) (h w) c -> b c t h w', b=b,h=h) + + if (no_temporal_attn is None) or (not no_temporal_attn): + if self.temporalcrossfirst: + # temporal cross attention + if self.temporal_crossattn_type == "crossattn": + # if temporal_context is not None: + temporal_context = context.repeat(h*w, 1, 1) + x = self.attn2_tmp(self.norm5(x), context=temporal_context, mask=mask) + x + elif self.temporal_crossattn_type == "selfattn": + x = self.attn2_tmp(self.norm5(x), context=None, mask=mask) + x + elif self.temporal_crossattn_type == "skip": + pass + else: + raise NotImplementedError + # temporal self attention + x = rearrange(x, 'b c t h w -> (b h w) t c') + x = self.attn1_tmp(self.norm4(x), mask=mask) + x + else: + # temporal self attention + x = rearrange(x, 'b c t h w -> (b h w) t c') + x = self.attn1_tmp(self.norm4(x), mask=mask) + x + # temporal cross attention + if self.temporal_crossattn_type == "crossattn": + if temporal_context is not None: + temporal_context = context.repeat(h*w, 1, 1) + x = self.attn2_tmp(self.norm5(x), context=temporal_context, mask=mask) + x + elif self.temporal_crossattn_type == "selfattn": + x = self.attn2_tmp(self.norm5(x), context=None, mask=mask) + x + elif self.temporal_crossattn_type == "skip": + pass + else: + raise NotImplementedError + else: + raise NotImplementedError + + return x + + def _st_cross_attn_parallel(self, x, context=None, temporal_context=None, order="sst", no_temporal_attn=None): + """ order: x -> Self Attn -> Cross Attn -> attn_s + x -> Temp Self Attn -> attn_t + x' = x + attn_s + attn_t + """ + if no_temporal_attn is not None: + raise NotImplementedError + + B, C, T, H, W = x.shape + # spatial self attention + h = x + h = rearrange(h, 'b c t h w -> (b t) (h w) c') + h = self.attn1(self.norm1(h)) + h + # spatial cross + # context_ = context.repeat(T, 1, 1) if context is not None else None + if context is not None: + context_ = [] + for i in range(context.shape[0]): + context_.append(context[i].unsqueeze(0).repeat(T, 1, 1)) + context_ = torch.cat(context_,dim=0) + else: + context_ = None + + h = self.attn2(self.norm2(h), context=context_) + h + h = rearrange(h, '(b t) (h w) c -> b c t h w', b=B, h=H) + + # temporal self + h2 = x + h2 = rearrange(h2, 'b c t h w -> (b h w) t c') + h2 = self.attn1_tmp(self.norm4(h2))# + h2 + h2 = rearrange(h2, '(b h w) t c -> b c t h w', b=B, h=H, w=W) + out = h + h2 + return rearrange(out, 'b c t h w -> (b h w) t c') + + ############################################################################################################################################## + +def spatial_attn_reshape(x): + return rearrange(x, 'b c t h w -> (b t) (h w) c') +def spatial_attn_reshape_back(x,b,h): + return rearrange(x, '(b t) (h w) c -> b c t h w', b=b,h=h) +def temporal_attn_reshape(x): + return rearrange(x, 'b c t h w -> (b h w) t c') +def temporal_attn_reshape_back(x, b,h,w): + return rearrange(x, '(b h w) t c -> b c t h w', b=b, h=h, w=w) +def local_spatial_temporal_attn_reshape(x, window_size): + B, C, T, H, W = x.shape + NH = H // window_size + NW = W // window_size + # x = x.view(B, C, T, NH, window_size, NW, window_size) + # tokens = x.permute(0, 1, 2, 3, 5, 4, 6).contiguous() + # tokens = tokens.view(-1, window_size, window_size, C) + x = rearrange(x, 'b c t (nh wh) (nw ww) -> b c t nh wh nw ww', nh=NH, nw=NW, wh=window_size, ww=window_size).contiguous() # # B, C, T, NH, NW, window_size, window_size + x = rearrange(x, 'b c t nh wh nw ww -> (b nh nw) (t wh ww) c') # (B, NH, NW) (T, window_size, window_size) C + return x +def local_spatial_temporal_attn_reshape_back(x, window_size, b, h, w, t): + B, L, C = x.shape + NH = h // window_size + NW = w // window_size + x = rearrange(x, '(b nh nw) (t wh ww) c -> b c t nh wh nw ww', b=b, nh=NH, nw=NW, t=t, wh=window_size, ww=window_size) + x = rearrange(x, 'b c t nh wh nw ww -> b c t (nh wh) (nw ww)') + return x + + +class SpatialTemporalTransformer(nn.Module): + """ + Transformer block for video-like data (5D tensor). + First, project the input (aka embedding) with NO reshape. + Then apply standard transformer action. + The 5D -> 3D reshape operation will be done in the specific attention module. + """ + def __init__( + self, + in_channels, n_heads, d_head, + depth=1, dropout=0., + context_dim=None, + # Temporal stuff + temporal_length=None, + image_length=None, + use_relative_position=True, + img_video_joint_train=False, + cross_attn_on_tempoal=False, + temporal_crossattn_type=False, + order="stst", + temporalcrossfirst=False, + split_stcontext=False, + temporal_context_dim=None, + **kwargs, + ): + super().__init__() + + self.in_channels = in_channels + inner_dim = n_heads * d_head + + self.norm = Normalize(in_channels) + self.proj_in = nn.Conv3d(in_channels, + inner_dim, + kernel_size=1, + stride=1, + padding=0) + + self.transformer_blocks = nn.ModuleList( + [BasicTransformerBlockST( + inner_dim, n_heads, d_head, dropout=dropout, + # cross attn + context_dim=context_dim, + # temporal attn + temporal_length=temporal_length, + image_length=image_length, + use_relative_position=use_relative_position, + img_video_joint_train=img_video_joint_train, + temporal_crossattn_type=temporal_crossattn_type, + order=order, + temporalcrossfirst=temporalcrossfirst, + split_stcontext=split_stcontext, + temporal_context_dim=temporal_context_dim, + **kwargs + ) for d in range(depth)] + ) + + self.proj_out = zero_module(nn.Conv3d(inner_dim, + in_channels, + kernel_size=1, + stride=1, + padding=0)) + + def forward(self, x, context=None, temporal_context=None, **kwargs): + # note: if no context is given, cross-attention defaults to self-attention + assert(x.dim() == 5), f"x shape = {x.shape}" + b, c, t, h, w = x.shape + x_in = x + + x = self.norm(x) + x = self.proj_in(x) + + for block in self.transformer_blocks: + x = block(x, context=context, temporal_context=temporal_context, **kwargs) + + x = self.proj_out(x) + return x + x_in + +# --------------------------------------------------------------------------------------------------- + +class STAttentionBlock2(nn.Module): + def __init__( + self, + channels, + num_heads=1, + num_head_channels=-1, + use_checkpoint=False, # not used, only used in ResBlock + use_new_attention_order=False, # QKVAttention or QKVAttentionLegacy + temporal_length=16, # used in relative positional representation. + image_length=8, # used for image-video joint training. + use_relative_position=False, # whether use relative positional representation in temporal attention. + img_video_joint_train=False, + # norm_type="groupnorm", + attn_norm_type="group", + use_tempoal_causal_attn=False, + ): + """ + version 1: guided_diffusion implemented version + version 2: remove args input argument + """ + super().__init__() + + if num_head_channels == -1: + self.num_heads = num_heads + else: + assert ( + channels % num_head_channels == 0 + ), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}" + self.num_heads = channels // num_head_channels + self.use_checkpoint = use_checkpoint + + self.temporal_length = temporal_length + self.image_length = image_length + self.use_relative_position = use_relative_position + self.img_video_joint_train = img_video_joint_train + self.attn_norm_type = attn_norm_type + assert(self.attn_norm_type in ["group", "no_norm"]) + self.use_tempoal_causal_attn = use_tempoal_causal_attn + + if self.attn_norm_type == "group": + self.norm_s = normalization(channels) + self.norm_t = normalization(channels) + + self.qkv_s = conv_nd(1, channels, channels * 3, 1) + self.qkv_t = conv_nd(1, channels, channels * 3, 1) + + if self.img_video_joint_train: + mask = th.ones([1, temporal_length+image_length, temporal_length+image_length]) + mask[:, temporal_length:, :] = 0 + mask[:, :, temporal_length:] = 0 + self.register_buffer("mask", mask) + else: + self.mask = None + + if use_new_attention_order: + # split qkv before split heads + self.attention_s = QKVAttention(self.num_heads) + self.attention_t = QKVAttention(self.num_heads) + else: + # split heads before split qkv + self.attention_s = QKVAttentionLegacy(self.num_heads) + self.attention_t = QKVAttentionLegacy(self.num_heads) + + if use_relative_position: + self.relative_position_k = RelativePosition(num_units=channels // self.num_heads, max_relative_position=temporal_length) + self.relative_position_v = RelativePosition(num_units=channels // self.num_heads, max_relative_position=temporal_length) + + self.proj_out_s = zero_module(conv_nd(1, channels, channels, 1)) # conv_dim, in_channels, out_channels, kernel_size + self.proj_out_t = zero_module(conv_nd(1, channels, channels, 1)) # conv_dim, in_channels, out_channels, kernel_size + + def forward(self, x, mask=None): + b, c, t, h, w = x.shape + + # spatial + out = rearrange(x, 'b c t h w -> (b t) c (h w)') + if self.attn_norm_type == "no_norm": + qkv = self.qkv_s(out) + else: + qkv = self.qkv_s(self.norm_s(out)) + out = self.attention_s(qkv) + out = self.proj_out_s(out) + out = rearrange(out, '(b t) c (h w) -> b c t h w', b=b,h=h) + x += out + + # temporal + out = rearrange(x, 'b c t h w -> (b h w) c t') + if self.attn_norm_type == "no_norm": + qkv = self.qkv_t(out) + else: + qkv = self.qkv_t(self.norm_t(out)) + + # relative positional embedding + if self.use_relative_position: + len_q = qkv.size()[-1] + len_k, len_v = len_q, len_q + k_rp = self.relative_position_k(len_q, len_k) + v_rp = self.relative_position_v(len_q, len_v) #[T,T,head_dim] + out = self.attention_t(qkv, rp=(k_rp, v_rp), mask=self.mask, use_tempoal_causal_attn=self.use_tempoal_causal_attn) + else: + out = self.attention_t(qkv, rp=None, mask=self.mask, use_tempoal_causal_attn=self.use_tempoal_causal_attn) + + out = self.proj_out_t(out) + out = rearrange(out, '(b h w) c t -> b c t h w', b=b,h=h,w=w) + + return (x + out) + +# --------------------------------------------------------------------------------------------------------------- + +class QKVAttentionLegacy(nn.Module): + """ + A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping + """ + + def __init__(self, n_heads): + super().__init__() + self.n_heads = n_heads + + def forward(self, qkv, rp=None, mask=None): + """ + Apply QKV attention. + + :param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs. + :return: an [N x (H * C) x T] tensor after attention. + """ + if rp is not None or mask is not None: + raise NotImplementedError + bs, width, length = qkv.shape + assert width % (3 * self.n_heads) == 0 + ch = width // (3 * self.n_heads) + q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1) + scale = 1 / math.sqrt(math.sqrt(ch)) + weight = th.einsum( + "bct,bcs->bts", q * scale, k * scale + ) # More stable with f16 than dividing afterwards + weight = th.softmax(weight.float(), dim=-1).type(weight.dtype) + a = th.einsum("bts,bcs->bct", weight, v) + return a.reshape(bs, -1, length) + + @staticmethod + def count_flops(model, _x, y): + return count_flops_attn(model, _x, y) + +# --------------------------------------------------------------------------------------------------------------- + +class QKVAttention(nn.Module): + """ + A module which performs QKV attention and splits in a different order. + """ + + def __init__(self, n_heads): + super().__init__() + self.n_heads = n_heads + + def forward(self, qkv, rp=None, mask=None, use_tempoal_causal_attn=False): + """ + Apply QKV attention. + + :param qkv: an [N x (3 * H * C) x T] tensor of Qs, Ks, and Vs. + :return: an [N x (H * C) x T] tensor after attention. + """ + bs, width, length = qkv.shape + assert width % (3 * self.n_heads) == 0 + ch = width // (3 * self.n_heads) + # print('qkv', qkv.size()) + q, k, v = qkv.chunk(3, dim=1) + scale = 1 / math.sqrt(math.sqrt(ch)) + # print('bs, self.n_heads, ch, length', bs, self.n_heads, ch, length) + + weight = th.einsum( + "bct,bcs->bts", + (q * scale).view(bs * self.n_heads, ch, length), + (k * scale).view(bs * self.n_heads, ch, length), + ) # More stable with f16 than dividing afterwards + # weight:[b,t,s] b=bs*n_heads*T + + if rp is not None: + k_rp, v_rp = rp # [length, length, head_dim] [8, 8, 48] + weight2 = th.einsum( + 'bct,tsc->bst', + (q * scale).view(bs * self.n_heads, ch, length), + k_rp + ) + weight += weight2 + + if use_tempoal_causal_attn: + # weight = torch.tril(weight) + assert(mask is None), f'Not implemented for merging two masks!' + mask = torch.tril(torch.ones(weight.shape)) + else: + if mask is not None: # only keep upper-left matrix + # process mask + c, t, _ = weight.shape + + if mask.shape[-1] > t: + mask = mask[:, :t, :t] + elif mask.shape[-1] < t: # pad ones + mask_ = th.zeros([c,t,t]).to(mask.device) + t_ = mask.shape[-1] + mask_[:, :t_, :t_] = mask + mask = mask_ + else: + assert(weight.shape[-1] == mask.shape[-1]), f'weight={weight.shape}, mask={mask.shape}' + + if mask is not None: + INF = -1e8 #float('-inf') + weight = weight.float().masked_fill(mask == 0, INF) + + weight = F.softmax(weight.float(), dim=-1).type(weight.dtype) #[256, 8, 8] [b, t, t] b=bs*n_heads*h*w,t=nframes + # weight = F.softmax(weight, dim=-1)#[256, 8, 8] [b, t, t] b=bs*n_heads*h*w,t=nframes + a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length)) #[256, 48, 8] [b, head_dim, t] + + if rp is not None: + a2 = th.einsum( + "bts,tsc->btc", + weight, + v_rp + ).transpose(1,2) # btc->bct + a += a2 + + return a.reshape(bs, -1, length) + +# --------------------------------------------------------------------------------------------------------------- + +# --------------------------------------------------------------------------------------------------------------- diff --git a/lvdm/modules/encoders/__init__.py b/lvdm/modules/encoders/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/lvdm/modules/encoders/adapter.py b/lvdm/modules/encoders/adapter.py new file mode 100644 index 0000000000000000000000000000000000000000..a65a1feae97853ef23cead5e2462c0e547392336 --- /dev/null +++ b/lvdm/modules/encoders/adapter.py @@ -0,0 +1,102 @@ +import torch.nn as nn + +from lvdm.basics import avg_pool_nd, conv_nd + + +class Downsample(nn.Module): + """ + A downsampling layer with an optional convolution. + :param channels: channels in the inputs and outputs. + :param use_conv: a bool determining if a convolution is applied. + :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then + downsampling occurs in the inner-two dimensions. + """ + + def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1): + super().__init__() + self.channels = channels + self.out_channels = out_channels or channels + self.use_conv = use_conv + self.dims = dims + stride = 2 if dims != 3 else (1, 2, 2) + if use_conv: + self.op = conv_nd( + dims, self.channels, self.out_channels, 3, stride=stride, padding=padding + ) + else: + assert self.channels == self.out_channels + self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride) + + def forward(self, x): + assert x.shape[1] == self.channels + return self.op(x) + + +class ResnetBlock(nn.Module): + def __init__(self, in_c, out_c, down, ksize=3, sk=False, use_conv=True): + super().__init__() + ps = ksize // 2 + if in_c != out_c or sk == False: + self.in_conv = nn.Conv2d(in_c, out_c, ksize, 1, ps) + else: + # print('n_in') + self.in_conv = None + self.block1 = nn.Conv2d(out_c, out_c, 3, 1, 1) + self.act = nn.ReLU() + self.block2 = nn.Conv2d(out_c, out_c, ksize, 1, ps) + if sk == False: + # self.skep = nn.Conv2d(in_c, out_c, ksize, 1, ps) # edit by zhouxiawang + self.skep = nn.Conv2d(out_c, out_c, ksize, 1, ps) + else: + self.skep = None + + self.down = down + if self.down == True: + self.down_opt = Downsample(in_c, use_conv=use_conv) + + def forward(self, x): + if self.down == True: + x = self.down_opt(x) + if self.in_conv is not None: # edit + x = self.in_conv(x) + + h = self.block1(x) + h = self.act(h) + h = self.block2(h) + if self.skep is not None: + return h + self.skep(x) + else: + return h + x + + +class Adapter(nn.Module): + def __init__(self, channels=[320, 640, 1280, 1280], nums_rb=3, cin=64, ksize=3, sk=False, use_conv=True): + super(Adapter, self).__init__() + self.unshuffle = nn.PixelUnshuffle(8) + self.channels = channels + self.nums_rb = nums_rb + self.body = [] + for i in range(len(channels)): + for j in range(nums_rb): + if (i != 0) and (j == 0): + self.body.append( + ResnetBlock(channels[i - 1], channels[i], down=True, ksize=ksize, sk=sk, use_conv=use_conv)) + else: + self.body.append( + ResnetBlock(channels[i], channels[i], down=False, ksize=ksize, sk=sk, use_conv=use_conv)) + self.body = nn.ModuleList(self.body) + self.conv_in = nn.Conv2d(cin, channels[0], 3, 1, 1) + + def forward(self, x): + # unshuffle + x = self.unshuffle(x) + # extract features + features = [] + x = self.conv_in(x) + for i in range(len(self.channels)): + for j in range(self.nums_rb): + idx = i * self.nums_rb + j + x = self.body[idx](x) + features.append(x) + + return features diff --git a/lvdm/modules/encoders/condition2.py b/lvdm/modules/encoders/condition2.py new file mode 100644 index 0000000000000000000000000000000000000000..d8e8eb97bbfae38961ce232a02319e768bbda440 --- /dev/null +++ b/lvdm/modules/encoders/condition2.py @@ -0,0 +1,352 @@ +import kornia +import open_clip +import torch +import torch.nn as nn +from torch.utils.checkpoint import checkpoint +from transformers import (CLIPTextModel, CLIPTokenizer, T5EncoderModel, + T5Tokenizer) + +from lvdm.common import autocast +from utils.utils import count_params + + +class AbstractEncoder(nn.Module): + def __init__(self): + super().__init__() + + def encode(self, *args, **kwargs): + raise NotImplementedError + + +class IdentityEncoder(AbstractEncoder): + + def encode(self, x): + return x + + +class ClassEmbedder(nn.Module): + def __init__(self, embed_dim, n_classes=1000, key='class', ucg_rate=0.1): + super().__init__() + self.key = key + self.embedding = nn.Embedding(n_classes, embed_dim) + self.n_classes = n_classes + self.ucg_rate = ucg_rate + + def forward(self, batch, key=None, disable_dropout=False): + if key is None: + key = self.key + # this is for use in crossattn + c = batch[key][:, None] + if self.ucg_rate > 0. and not disable_dropout: + mask = 1. - torch.bernoulli(torch.ones_like(c) * self.ucg_rate) + c = mask * c + (1 - mask) * torch.ones_like(c) * (self.n_classes - 1) + c = c.long() + c = self.embedding(c) + return c + + def get_unconditional_conditioning(self, bs, device="cuda"): + uc_class = self.n_classes - 1 # 1000 classes --> 0 ... 999, one extra class for ucg (class 1000) + uc = torch.ones((bs,), device=device) * uc_class + uc = {self.key: uc} + return uc + + +def disabled_train(self, mode=True): + """Overwrite model.train with this function to make sure train/eval mode + does not change anymore.""" + return self + + +class FrozenT5Embedder(AbstractEncoder): + """Uses the T5 transformer encoder for text""" + + def __init__(self, version="google/t5-v1_1-large", device="cuda", max_length=77, + freeze=True): # others are google/t5-v1_1-xl and google/t5-v1_1-xxl + super().__init__() + self.tokenizer = T5Tokenizer.from_pretrained(version) + self.transformer = T5EncoderModel.from_pretrained(version) + self.device = device + self.max_length = max_length # TODO: typical value? + if freeze: + self.freeze() + + def freeze(self): + self.transformer = self.transformer.eval() + # self.train = disabled_train + for param in self.parameters(): + param.requires_grad = False + + def forward(self, text): + batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True, + return_overflowing_tokens=False, padding="max_length", return_tensors="pt") + tokens = batch_encoding["input_ids"].to(self.device) + outputs = self.transformer(input_ids=tokens) + + z = outputs.last_hidden_state + return z + + def encode(self, text): + return self(text) + + +class FrozenCLIPEmbedder(AbstractEncoder): + """Uses the CLIP transformer encoder for text (from huggingface)""" + LAYERS = [ + "last", + "pooled", + "hidden" + ] + + def __init__(self, version="openai/clip-vit-large-patch14", device="cuda", max_length=77, + freeze=True, layer="last", layer_idx=None): # clip-vit-base-patch32 + super().__init__() + assert layer in self.LAYERS + self.tokenizer = CLIPTokenizer.from_pretrained(version) + self.transformer = CLIPTextModel.from_pretrained(version) + self.device = device + self.max_length = max_length + if freeze: + self.freeze() + self.layer = layer + self.layer_idx = layer_idx + if layer == "hidden": + assert layer_idx is not None + assert 0 <= abs(layer_idx) <= 12 + + def freeze(self): + self.transformer = self.transformer.eval() + # self.train = disabled_train + for param in self.parameters(): + param.requires_grad = False + + def forward(self, text): + batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True, + return_overflowing_tokens=False, padding="max_length", return_tensors="pt") + tokens = batch_encoding["input_ids"].to(self.device) + outputs = self.transformer(input_ids=tokens, output_hidden_states=self.layer == "hidden") + if self.layer == "last": + z = outputs.last_hidden_state + elif self.layer == "pooled": + z = outputs.pooler_output[:, None, :] + else: + z = outputs.hidden_states[self.layer_idx] + return z + + def encode(self, text): + return self(text) + + +class ClipImageEmbedder(nn.Module): + def __init__( + self, + model, + jit=False, + device='cuda' if torch.cuda.is_available() else 'cpu', + antialias=True, + ucg_rate=0. + ): + super().__init__() + from clip import load as load_clip + self.model, _ = load_clip(name=model, device=device, jit=jit) + + self.antialias = antialias + + self.register_buffer('mean', torch.Tensor([0.48145466, 0.4578275, 0.40821073]), persistent=False) + self.register_buffer('std', torch.Tensor([0.26862954, 0.26130258, 0.27577711]), persistent=False) + self.ucg_rate = ucg_rate + + def preprocess(self, x): + # normalize to [0,1] + x = kornia.geometry.resize(x, (224, 224), + interpolation='bicubic', align_corners=True, + antialias=self.antialias) + x = (x + 1.) / 2. + # re-normalize according to clip + x = kornia.enhance.normalize(x, self.mean, self.std) + return x + + def forward(self, x, no_dropout=False): + # x is assumed to be in range [-1,1] + out = self.model.encode_image(self.preprocess(x)) + out = out.to(x.dtype) + if self.ucg_rate > 0. and not no_dropout: + out = torch.bernoulli((1. - self.ucg_rate) * torch.ones(out.shape[0], device=out.device))[:, None] * out + return out + + +class FrozenOpenCLIPEmbedder(AbstractEncoder): + """ + Uses the OpenCLIP transformer encoder for text + """ + LAYERS = [ + # "pooled", + "last", + "penultimate" + ] + + def __init__(self, arch="ViT-H-14", version="laion2b_s32b_b79k", device="cuda", max_length=77, + freeze=True, layer="last"): + super().__init__() + assert layer in self.LAYERS + # model, _, _ = open_clip.create_model_and_transforms(arch, device=torch.device('cpu'), pretrained='/apdcephfs/share_1290939/richardxia/PretrainedCache/hub/models--laion--CLIP-ViT-H-14-laion2B-s32B-b79K/snapshots/719803079cc9d41bf3ad0a0916fa24e778320c50/open_clip_pytorch_model.bin') + model, _, _ = open_clip.create_model_and_transforms('hf-hub:laion/CLIP-ViT-H-14-laion2B-s32B-b79K') + del model.visual + self.model = model + + self.device = device + self.max_length = max_length + if freeze: + self.freeze() + self.layer = layer + if self.layer == "last": + self.layer_idx = 0 + elif self.layer == "penultimate": + self.layer_idx = 1 + else: + raise NotImplementedError() + + def freeze(self): + self.model = self.model.eval() + for param in self.parameters(): + param.requires_grad = False + + def forward(self, text): + tokens = open_clip.tokenize(text) + z = self.encode_with_transformer(tokens.to(self.device)) + return z + + def encode_with_transformer(self, text): + x = self.model.token_embedding(text) # [batch_size, n_ctx, d_model] + x = x + self.model.positional_embedding + x = x.permute(1, 0, 2) # NLD -> LND + x = self.text_transformer_forward(x, attn_mask=self.model.attn_mask) + x = x.permute(1, 0, 2) # LND -> NLD + x = self.model.ln_final(x) + return x + + def text_transformer_forward(self, x: torch.Tensor, attn_mask=None): + for i, r in enumerate(self.model.transformer.resblocks): + if i == len(self.model.transformer.resblocks) - self.layer_idx: + break + if self.model.transformer.grad_checkpointing and not torch.jit.is_scripting(): + x = checkpoint(r, x, attn_mask) + else: + x = r(x, attn_mask=attn_mask) + return x + + def encode(self, text): + return self(text) + + +class FrozenOpenCLIPImageEmbedder(AbstractEncoder): + """ + Uses the OpenCLIP vision transformer encoder for images + """ + + def __init__(self, arch="ViT-H-14", version="laion2b_s32b_b79k", device="cuda", max_length=77, + freeze=True, layer="pooled", antialias=True, ucg_rate=0.): + super().__init__() + model, _, _ = open_clip.create_model_and_transforms(arch, device=torch.device('cpu'), + pretrained=version, ) + del model.transformer + self.model = model + + self.device = device + self.max_length = max_length + if freeze: + self.freeze() + self.layer = layer + if self.layer == "penultimate": + raise NotImplementedError() + self.layer_idx = 1 + + self.antialias = antialias + + self.register_buffer('mean', torch.Tensor([0.48145466, 0.4578275, 0.40821073]), persistent=False) + self.register_buffer('std', torch.Tensor([0.26862954, 0.26130258, 0.27577711]), persistent=False) + self.ucg_rate = ucg_rate + + def preprocess(self, x): + # normalize to [0,1] + x = kornia.geometry.resize(x, (224, 224), + interpolation='bicubic', align_corners=True, + antialias=self.antialias) + x = (x + 1.) / 2. + # renormalize according to clip + x = kornia.enhance.normalize(x, self.mean, self.std) + return x + + def freeze(self): + self.model = self.model.eval() + for param in self.parameters(): + param.requires_grad = False + + @autocast + def forward(self, image, no_dropout=False): + z = self.encode_with_vision_transformer(image) + if self.ucg_rate > 0. and not no_dropout: + z = torch.bernoulli((1. - self.ucg_rate) * torch.ones(z.shape[0], device=z.device))[:, None] * z + return z + + def encode_with_vision_transformer(self, img): + img = self.preprocess(img) + x = self.model.visual(img) + return x + + def encode(self, text): + return self(text) + + +class FrozenCLIPT5Encoder(AbstractEncoder): + def __init__(self, clip_version="openai/clip-vit-large-patch14", t5_version="google/t5-v1_1-xl", device="cuda", + clip_max_length=77, t5_max_length=77): + super().__init__() + self.clip_encoder = FrozenCLIPEmbedder(clip_version, device, max_length=clip_max_length) + self.t5_encoder = FrozenT5Embedder(t5_version, device, max_length=t5_max_length) + print(f"{self.clip_encoder.__class__.__name__} has {count_params(self.clip_encoder) * 1.e-6:.2f} M parameters, " + f"{self.t5_encoder.__class__.__name__} comes with {count_params(self.t5_encoder) * 1.e-6:.2f} M params.") + + def encode(self, text): + return self(text) + + def forward(self, text): + clip_z = self.clip_encoder.encode(text) + t5_z = self.t5_encoder.encode(text) + return [clip_z, t5_z] + +''' +from ldm.modules.diffusionmodules.upscaling import ImageConcatWithNoiseAugmentation +from ldm.modules.diffusionmodules.openaimodel import Timestep + +class CLIPEmbeddingNoiseAugmentation(ImageConcatWithNoiseAugmentation): + def __init__(self, *args, clip_stats_path=None, timestep_dim=256, **kwargs): + super().__init__(*args, **kwargs) + if clip_stats_path is None: + clip_mean, clip_std = torch.zeros(timestep_dim), torch.ones(timestep_dim) + else: + clip_mean, clip_std = torch.load(clip_stats_path, map_location="cpu") + self.register_buffer("data_mean", clip_mean[None, :], persistent=False) + self.register_buffer("data_std", clip_std[None, :], persistent=False) + self.time_embed = Timestep(timestep_dim) + + def scale(self, x): + # re-normalize to centered mean and unit variance + x = (x - self.data_mean) * 1. / self.data_std + return x + + def unscale(self, x): + # back to original data stats + x = (x * self.data_std) + self.data_mean + return x + + def forward(self, x, noise_level=None): + if noise_level is None: + noise_level = torch.randint(0, self.max_noise_level, (x.shape[0],), device=x.device).long() + else: + assert isinstance(noise_level, torch.Tensor) + x = self.scale(x) + z = self.q_sample(x, noise_level) + z = self.unscale(z) + noise_level = self.time_embed(noise_level) + return z, noise_level +''' \ No newline at end of file diff --git a/lvdm/modules/networks/ae_modules.py b/lvdm/modules/networks/ae_modules.py new file mode 100644 index 0000000000000000000000000000000000000000..034dd0aa5c0bcaccbfb05462748050fd2fee67aa --- /dev/null +++ b/lvdm/modules/networks/ae_modules.py @@ -0,0 +1,847 @@ +# pytorch_diffusion + derived encoder decoder +import math + +import torch +import numpy as np +import torch.nn as nn +from einops import rearrange + +from utils.utils import instantiate_from_config +from lvdm.modules.attention import LinearAttention + +def nonlinearity(x): + # swish + return x*torch.sigmoid(x) + + +def Normalize(in_channels, num_groups=32): + return torch.nn.GroupNorm(num_groups=num_groups, num_channels=in_channels, eps=1e-6, affine=True) + + + +class LinAttnBlock(LinearAttention): + """to match AttnBlock usage""" + def __init__(self, in_channels): + super().__init__(dim=in_channels, heads=1, dim_head=in_channels) + + +class AttnBlock(nn.Module): + def __init__(self, in_channels): + super().__init__() + self.in_channels = in_channels + + self.norm = Normalize(in_channels) + self.q = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0) + self.k = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0) + self.v = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0) + self.proj_out = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=1, + stride=1, + padding=0) + + def forward(self, x): + h_ = x + h_ = self.norm(h_) + q = self.q(h_) + k = self.k(h_) + v = self.v(h_) + + # compute attention + b,c,h,w = q.shape + q = q.reshape(b,c,h*w) # bcl + q = q.permute(0,2,1) # bcl -> blc l=hw + k = k.reshape(b,c,h*w) # bcl + + w_ = torch.bmm(q,k) # b,hw,hw w[b,i,j]=sum_c q[b,i,c]k[b,c,j] + w_ = w_ * (int(c)**(-0.5)) + w_ = torch.nn.functional.softmax(w_, dim=2) + + # attend to values + v = v.reshape(b,c,h*w) + w_ = w_.permute(0,2,1) # b,hw,hw (first hw of k, second of q) + h_ = torch.bmm(v,w_) # b, c,hw (hw of q) h_[b,c,j] = sum_i v[b,c,i] w_[b,i,j] + h_ = h_.reshape(b,c,h,w) + + h_ = self.proj_out(h_) + + return x+h_ + +def make_attn(in_channels, attn_type="vanilla"): + assert attn_type in ["vanilla", "linear", "none"], f'attn_type {attn_type} unknown' + #print(f"making attention of type '{attn_type}' with {in_channels} in_channels") + if attn_type == "vanilla": + return AttnBlock(in_channels) + elif attn_type == "none": + return nn.Identity(in_channels) + else: + return LinAttnBlock(in_channels) + +class Downsample(nn.Module): + def __init__(self, in_channels, with_conv): + super().__init__() + self.with_conv = with_conv + self.in_channels = in_channels + if self.with_conv: + # no asymmetric padding in torch conv, must do it ourselves + self.conv = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=3, + stride=2, + padding=0) + def forward(self, x): + if self.with_conv: + pad = (0,1,0,1) + x = torch.nn.functional.pad(x, pad, mode="constant", value=0) + x = self.conv(x) + else: + x = torch.nn.functional.avg_pool2d(x, kernel_size=2, stride=2) + return x + +class Upsample(nn.Module): + def __init__(self, in_channels, with_conv): + super().__init__() + self.with_conv = with_conv + self.in_channels = in_channels + if self.with_conv: + self.conv = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=3, + stride=1, + padding=1) + + def forward(self, x): + x = torch.nn.functional.interpolate(x, scale_factor=2.0, mode="nearest") + if self.with_conv: + x = self.conv(x) + return x + +def get_timestep_embedding(timesteps, embedding_dim): + """ + This matches the implementation in Denoising Diffusion Probabilistic Models: + From Fairseq. + Build sinusoidal embeddings. + This matches the implementation in tensor2tensor, but differs slightly + from the description in Section 3.5 of "Attention Is All You Need". + """ + assert len(timesteps.shape) == 1 + + half_dim = embedding_dim // 2 + emb = math.log(10000) / (half_dim - 1) + emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb) + emb = emb.to(device=timesteps.device) + emb = timesteps.float()[:, None] * emb[None, :] + emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1) + if embedding_dim % 2 == 1: # zero pad + emb = torch.nn.functional.pad(emb, (0,1,0,0)) + return emb + + + +class ResnetBlock(nn.Module): + def __init__(self, *, in_channels, out_channels=None, conv_shortcut=False, + dropout, temb_channels=512): + super().__init__() + self.in_channels = in_channels + out_channels = in_channels if out_channels is None else out_channels + self.out_channels = out_channels + self.use_conv_shortcut = conv_shortcut + + self.norm1 = Normalize(in_channels) + self.conv1 = torch.nn.Conv2d(in_channels, + out_channels, + kernel_size=3, + stride=1, + padding=1) + if temb_channels > 0: + self.temb_proj = torch.nn.Linear(temb_channels, + out_channels) + self.norm2 = Normalize(out_channels) + self.dropout = torch.nn.Dropout(dropout) + self.conv2 = torch.nn.Conv2d(out_channels, + out_channels, + kernel_size=3, + stride=1, + padding=1) + if self.in_channels != self.out_channels: + if self.use_conv_shortcut: + self.conv_shortcut = torch.nn.Conv2d(in_channels, + out_channels, + kernel_size=3, + stride=1, + padding=1) + else: + self.nin_shortcut = torch.nn.Conv2d(in_channels, + out_channels, + kernel_size=1, + stride=1, + padding=0) + + def forward(self, x, temb): + h = x + h = self.norm1(h) + h = nonlinearity(h) + h = self.conv1(h) + + if temb is not None: + h = h + self.temb_proj(nonlinearity(temb))[:,:,None,None] + + h = self.norm2(h) + h = nonlinearity(h) + h = self.dropout(h) + h = self.conv2(h) + + if self.in_channels != self.out_channels: + if self.use_conv_shortcut: + x = self.conv_shortcut(x) + else: + x = self.nin_shortcut(x) + + return x+h + +class Model(nn.Module): + def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks, + attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels, + resolution, use_timestep=True, use_linear_attn=False, attn_type="vanilla"): + super().__init__() + if use_linear_attn: attn_type = "linear" + self.ch = ch + self.temb_ch = self.ch*4 + self.num_resolutions = len(ch_mult) + self.num_res_blocks = num_res_blocks + self.resolution = resolution + self.in_channels = in_channels + + self.use_timestep = use_timestep + if self.use_timestep: + # timestep embedding + self.temb = nn.Module() + self.temb.dense = nn.ModuleList([ + torch.nn.Linear(self.ch, + self.temb_ch), + torch.nn.Linear(self.temb_ch, + self.temb_ch), + ]) + + # downsampling + self.conv_in = torch.nn.Conv2d(in_channels, + self.ch, + kernel_size=3, + stride=1, + padding=1) + + curr_res = resolution + in_ch_mult = (1,)+tuple(ch_mult) + self.down = nn.ModuleList() + for i_level in range(self.num_resolutions): + block = nn.ModuleList() + attn = nn.ModuleList() + block_in = ch*in_ch_mult[i_level] + block_out = ch*ch_mult[i_level] + for i_block in range(self.num_res_blocks): + block.append(ResnetBlock(in_channels=block_in, + out_channels=block_out, + temb_channels=self.temb_ch, + dropout=dropout)) + block_in = block_out + if curr_res in attn_resolutions: + attn.append(make_attn(block_in, attn_type=attn_type)) + down = nn.Module() + down.block = block + down.attn = attn + if i_level != self.num_resolutions-1: + down.downsample = Downsample(block_in, resamp_with_conv) + curr_res = curr_res // 2 + self.down.append(down) + + # middle + self.mid = nn.Module() + self.mid.block_1 = ResnetBlock(in_channels=block_in, + out_channels=block_in, + temb_channels=self.temb_ch, + dropout=dropout) + self.mid.attn_1 = make_attn(block_in, attn_type=attn_type) + self.mid.block_2 = ResnetBlock(in_channels=block_in, + out_channels=block_in, + temb_channels=self.temb_ch, + dropout=dropout) + + # upsampling + self.up = nn.ModuleList() + for i_level in reversed(range(self.num_resolutions)): + block = nn.ModuleList() + attn = nn.ModuleList() + block_out = ch*ch_mult[i_level] + skip_in = ch*ch_mult[i_level] + for i_block in range(self.num_res_blocks+1): + if i_block == self.num_res_blocks: + skip_in = ch*in_ch_mult[i_level] + block.append(ResnetBlock(in_channels=block_in+skip_in, + out_channels=block_out, + temb_channels=self.temb_ch, + dropout=dropout)) + block_in = block_out + if curr_res in attn_resolutions: + attn.append(make_attn(block_in, attn_type=attn_type)) + up = nn.Module() + up.block = block + up.attn = attn + if i_level != 0: + up.upsample = Upsample(block_in, resamp_with_conv) + curr_res = curr_res * 2 + self.up.insert(0, up) # prepend to get consistent order + + # end + self.norm_out = Normalize(block_in) + self.conv_out = torch.nn.Conv2d(block_in, + out_ch, + kernel_size=3, + stride=1, + padding=1) + + def forward(self, x, t=None, context=None): + #assert x.shape[2] == x.shape[3] == self.resolution + if context is not None: + # assume aligned context, cat along channel axis + x = torch.cat((x, context), dim=1) + if self.use_timestep: + # timestep embedding + assert t is not None + temb = get_timestep_embedding(t, self.ch) + temb = self.temb.dense[0](temb) + temb = nonlinearity(temb) + temb = self.temb.dense[1](temb) + else: + temb = None + + # downsampling + hs = [self.conv_in(x)] + for i_level in range(self.num_resolutions): + for i_block in range(self.num_res_blocks): + h = self.down[i_level].block[i_block](hs[-1], temb) + if len(self.down[i_level].attn) > 0: + h = self.down[i_level].attn[i_block](h) + hs.append(h) + if i_level != self.num_resolutions-1: + hs.append(self.down[i_level].downsample(hs[-1])) + + # middle + h = hs[-1] + h = self.mid.block_1(h, temb) + h = self.mid.attn_1(h) + h = self.mid.block_2(h, temb) + + # upsampling + for i_level in reversed(range(self.num_resolutions)): + for i_block in range(self.num_res_blocks+1): + h = self.up[i_level].block[i_block]( + torch.cat([h, hs.pop()], dim=1), temb) + if len(self.up[i_level].attn) > 0: + h = self.up[i_level].attn[i_block](h) + if i_level != 0: + h = self.up[i_level].upsample(h) + + # end + h = self.norm_out(h) + h = nonlinearity(h) + h = self.conv_out(h) + return h + + def get_last_layer(self): + return self.conv_out.weight + + +class Encoder(nn.Module): + def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks, + attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels, + resolution, z_channels, double_z=True, use_linear_attn=False, attn_type="vanilla", + **ignore_kwargs): + super().__init__() + if use_linear_attn: attn_type = "linear" + self.ch = ch + self.temb_ch = 0 + self.num_resolutions = len(ch_mult) + self.num_res_blocks = num_res_blocks + self.resolution = resolution + self.in_channels = in_channels + + # downsampling + self.conv_in = torch.nn.Conv2d(in_channels, + self.ch, + kernel_size=3, + stride=1, + padding=1) + + curr_res = resolution + in_ch_mult = (1,)+tuple(ch_mult) + self.in_ch_mult = in_ch_mult + self.down = nn.ModuleList() + for i_level in range(self.num_resolutions): + block = nn.ModuleList() + attn = nn.ModuleList() + block_in = ch*in_ch_mult[i_level] + block_out = ch*ch_mult[i_level] + for i_block in range(self.num_res_blocks): + block.append(ResnetBlock(in_channels=block_in, + out_channels=block_out, + temb_channels=self.temb_ch, + dropout=dropout)) + block_in = block_out + if curr_res in attn_resolutions: + attn.append(make_attn(block_in, attn_type=attn_type)) + down = nn.Module() + down.block = block + down.attn = attn + if i_level != self.num_resolutions-1: + down.downsample = Downsample(block_in, resamp_with_conv) + curr_res = curr_res // 2 + self.down.append(down) + + # middle + self.mid = nn.Module() + self.mid.block_1 = ResnetBlock(in_channels=block_in, + out_channels=block_in, + temb_channels=self.temb_ch, + dropout=dropout) + self.mid.attn_1 = make_attn(block_in, attn_type=attn_type) + self.mid.block_2 = ResnetBlock(in_channels=block_in, + out_channels=block_in, + temb_channels=self.temb_ch, + dropout=dropout) + + # end + self.norm_out = Normalize(block_in) + self.conv_out = torch.nn.Conv2d(block_in, + 2*z_channels if double_z else z_channels, + kernel_size=3, + stride=1, + padding=1) + + def forward(self, x): + # timestep embedding + temb = None + + # print(f'encoder-input={x.shape}') + # downsampling + hs = [self.conv_in(x)] + # print(f'encoder-conv in feat={hs[0].shape}') + for i_level in range(self.num_resolutions): + for i_block in range(self.num_res_blocks): + h = self.down[i_level].block[i_block](hs[-1], temb) + # print(f'encoder-down feat={h.shape}') + if len(self.down[i_level].attn) > 0: + h = self.down[i_level].attn[i_block](h) + hs.append(h) + if i_level != self.num_resolutions-1: + # print(f'encoder-downsample (input)={hs[-1].shape}') + hs.append(self.down[i_level].downsample(hs[-1])) + # print(f'encoder-downsample (output)={hs[-1].shape}') + + # middle + h = hs[-1] + h = self.mid.block_1(h, temb) + # print(f'encoder-mid1 feat={h.shape}') + h = self.mid.attn_1(h) + h = self.mid.block_2(h, temb) + # print(f'encoder-mid2 feat={h.shape}') + + # end + h = self.norm_out(h) + h = nonlinearity(h) + h = self.conv_out(h) + # print(f'end feat={h.shape}') + return h + + +class Decoder(nn.Module): + def __init__(self, *, ch, out_ch, ch_mult=(1,2,4,8), num_res_blocks, + attn_resolutions, dropout=0.0, resamp_with_conv=True, in_channels, + resolution, z_channels, give_pre_end=False, tanh_out=False, use_linear_attn=False, + attn_type="vanilla", **ignorekwargs): + super().__init__() + if use_linear_attn: attn_type = "linear" + self.ch = ch + self.temb_ch = 0 + self.num_resolutions = len(ch_mult) + self.num_res_blocks = num_res_blocks + self.resolution = resolution + self.in_channels = in_channels + self.give_pre_end = give_pre_end + self.tanh_out = tanh_out + + # compute in_ch_mult, block_in and curr_res at lowest res + in_ch_mult = (1,)+tuple(ch_mult) + block_in = ch*ch_mult[self.num_resolutions-1] + curr_res = resolution // 2**(self.num_resolutions-1) + self.z_shape = (1,z_channels,curr_res,curr_res) + print("AE working on z of shape {} = {} dimensions.".format( + self.z_shape, np.prod(self.z_shape))) + + # z to block_in + self.conv_in = torch.nn.Conv2d(z_channels, + block_in, + kernel_size=3, + stride=1, + padding=1) + + # middle + self.mid = nn.Module() + self.mid.block_1 = ResnetBlock(in_channels=block_in, + out_channels=block_in, + temb_channels=self.temb_ch, + dropout=dropout) + self.mid.attn_1 = make_attn(block_in, attn_type=attn_type) + self.mid.block_2 = ResnetBlock(in_channels=block_in, + out_channels=block_in, + temb_channels=self.temb_ch, + dropout=dropout) + + # upsampling + self.up = nn.ModuleList() + for i_level in reversed(range(self.num_resolutions)): + block = nn.ModuleList() + attn = nn.ModuleList() + block_out = ch*ch_mult[i_level] + for i_block in range(self.num_res_blocks+1): + block.append(ResnetBlock(in_channels=block_in, + out_channels=block_out, + temb_channels=self.temb_ch, + dropout=dropout)) + block_in = block_out + if curr_res in attn_resolutions: + attn.append(make_attn(block_in, attn_type=attn_type)) + up = nn.Module() + up.block = block + up.attn = attn + if i_level != 0: + up.upsample = Upsample(block_in, resamp_with_conv) + curr_res = curr_res * 2 + self.up.insert(0, up) # prepend to get consistent order + + # end + self.norm_out = Normalize(block_in) + self.conv_out = torch.nn.Conv2d(block_in, + out_ch, + kernel_size=3, + stride=1, + padding=1) + + def forward(self, z): + #assert z.shape[1:] == self.z_shape[1:] + self.last_z_shape = z.shape + + # print(f'decoder-input={z.shape}') + # timestep embedding + temb = None + + # z to block_in + h = self.conv_in(z) + # print(f'decoder-conv in feat={h.shape}') + + # middle + h = self.mid.block_1(h, temb) + h = self.mid.attn_1(h) + h = self.mid.block_2(h, temb) + # print(f'decoder-mid feat={h.shape}') + + # upsampling + for i_level in reversed(range(self.num_resolutions)): + for i_block in range(self.num_res_blocks+1): + h = self.up[i_level].block[i_block](h, temb) + if len(self.up[i_level].attn) > 0: + h = self.up[i_level].attn[i_block](h) + # print(f'decoder-up feat={h.shape}') + if i_level != 0: + h = self.up[i_level].upsample(h) + # print(f'decoder-upsample feat={h.shape}') + + # end + if self.give_pre_end: + return h + + h = self.norm_out(h) + h = nonlinearity(h) + h = self.conv_out(h) + # print(f'decoder-conv_out feat={h.shape}') + if self.tanh_out: + h = torch.tanh(h) + return h + + +class SimpleDecoder(nn.Module): + def __init__(self, in_channels, out_channels, *args, **kwargs): + super().__init__() + self.model = nn.ModuleList([nn.Conv2d(in_channels, in_channels, 1), + ResnetBlock(in_channels=in_channels, + out_channels=2 * in_channels, + temb_channels=0, dropout=0.0), + ResnetBlock(in_channels=2 * in_channels, + out_channels=4 * in_channels, + temb_channels=0, dropout=0.0), + ResnetBlock(in_channels=4 * in_channels, + out_channels=2 * in_channels, + temb_channels=0, dropout=0.0), + nn.Conv2d(2*in_channels, in_channels, 1), + Upsample(in_channels, with_conv=True)]) + # end + self.norm_out = Normalize(in_channels) + self.conv_out = torch.nn.Conv2d(in_channels, + out_channels, + kernel_size=3, + stride=1, + padding=1) + + def forward(self, x): + for i, layer in enumerate(self.model): + if i in [1,2,3]: + x = layer(x, None) + else: + x = layer(x) + + h = self.norm_out(x) + h = nonlinearity(h) + x = self.conv_out(h) + return x + + +class UpsampleDecoder(nn.Module): + def __init__(self, in_channels, out_channels, ch, num_res_blocks, resolution, + ch_mult=(2,2), dropout=0.0): + super().__init__() + # upsampling + self.temb_ch = 0 + self.num_resolutions = len(ch_mult) + self.num_res_blocks = num_res_blocks + block_in = in_channels + curr_res = resolution // 2 ** (self.num_resolutions - 1) + self.res_blocks = nn.ModuleList() + self.upsample_blocks = nn.ModuleList() + for i_level in range(self.num_resolutions): + res_block = [] + block_out = ch * ch_mult[i_level] + for i_block in range(self.num_res_blocks + 1): + res_block.append(ResnetBlock(in_channels=block_in, + out_channels=block_out, + temb_channels=self.temb_ch, + dropout=dropout)) + block_in = block_out + self.res_blocks.append(nn.ModuleList(res_block)) + if i_level != self.num_resolutions - 1: + self.upsample_blocks.append(Upsample(block_in, True)) + curr_res = curr_res * 2 + + # end + self.norm_out = Normalize(block_in) + self.conv_out = torch.nn.Conv2d(block_in, + out_channels, + kernel_size=3, + stride=1, + padding=1) + + def forward(self, x): + # upsampling + h = x + for k, i_level in enumerate(range(self.num_resolutions)): + for i_block in range(self.num_res_blocks + 1): + h = self.res_blocks[i_level][i_block](h, None) + if i_level != self.num_resolutions - 1: + h = self.upsample_blocks[k](h) + h = self.norm_out(h) + h = nonlinearity(h) + h = self.conv_out(h) + return h + + +class LatentRescaler(nn.Module): + def __init__(self, factor, in_channels, mid_channels, out_channels, depth=2): + super().__init__() + # residual block, interpolate, residual block + self.factor = factor + self.conv_in = nn.Conv2d(in_channels, + mid_channels, + kernel_size=3, + stride=1, + padding=1) + self.res_block1 = nn.ModuleList([ResnetBlock(in_channels=mid_channels, + out_channels=mid_channels, + temb_channels=0, + dropout=0.0) for _ in range(depth)]) + self.attn = AttnBlock(mid_channels) + self.res_block2 = nn.ModuleList([ResnetBlock(in_channels=mid_channels, + out_channels=mid_channels, + temb_channels=0, + dropout=0.0) for _ in range(depth)]) + + self.conv_out = nn.Conv2d(mid_channels, + out_channels, + kernel_size=1, + ) + + def forward(self, x): + x = self.conv_in(x) + for block in self.res_block1: + x = block(x, None) + x = torch.nn.functional.interpolate(x, size=(int(round(x.shape[2]*self.factor)), int(round(x.shape[3]*self.factor)))) + x = self.attn(x) + for block in self.res_block2: + x = block(x, None) + x = self.conv_out(x) + return x + + +class MergedRescaleEncoder(nn.Module): + def __init__(self, in_channels, ch, resolution, out_ch, num_res_blocks, + attn_resolutions, dropout=0.0, resamp_with_conv=True, + ch_mult=(1,2,4,8), rescale_factor=1.0, rescale_module_depth=1): + super().__init__() + intermediate_chn = ch * ch_mult[-1] + self.encoder = Encoder(in_channels=in_channels, num_res_blocks=num_res_blocks, ch=ch, ch_mult=ch_mult, + z_channels=intermediate_chn, double_z=False, resolution=resolution, + attn_resolutions=attn_resolutions, dropout=dropout, resamp_with_conv=resamp_with_conv, + out_ch=None) + self.rescaler = LatentRescaler(factor=rescale_factor, in_channels=intermediate_chn, + mid_channels=intermediate_chn, out_channels=out_ch, depth=rescale_module_depth) + + def forward(self, x): + x = self.encoder(x) + x = self.rescaler(x) + return x + + +class MergedRescaleDecoder(nn.Module): + def __init__(self, z_channels, out_ch, resolution, num_res_blocks, attn_resolutions, ch, ch_mult=(1,2,4,8), + dropout=0.0, resamp_with_conv=True, rescale_factor=1.0, rescale_module_depth=1): + super().__init__() + tmp_chn = z_channels*ch_mult[-1] + self.decoder = Decoder(out_ch=out_ch, z_channels=tmp_chn, attn_resolutions=attn_resolutions, dropout=dropout, + resamp_with_conv=resamp_with_conv, in_channels=None, num_res_blocks=num_res_blocks, + ch_mult=ch_mult, resolution=resolution, ch=ch) + self.rescaler = LatentRescaler(factor=rescale_factor, in_channels=z_channels, mid_channels=tmp_chn, + out_channels=tmp_chn, depth=rescale_module_depth) + + def forward(self, x): + x = self.rescaler(x) + x = self.decoder(x) + return x + + +class Upsampler(nn.Module): + def __init__(self, in_size, out_size, in_channels, out_channels, ch_mult=2): + super().__init__() + assert out_size >= in_size + num_blocks = int(np.log2(out_size//in_size))+1 + factor_up = 1.+ (out_size % in_size) + print(f"Building {self.__class__.__name__} with in_size: {in_size} --> out_size {out_size} and factor {factor_up}") + self.rescaler = LatentRescaler(factor=factor_up, in_channels=in_channels, mid_channels=2*in_channels, + out_channels=in_channels) + self.decoder = Decoder(out_ch=out_channels, resolution=out_size, z_channels=in_channels, num_res_blocks=2, + attn_resolutions=[], in_channels=None, ch=in_channels, + ch_mult=[ch_mult for _ in range(num_blocks)]) + + def forward(self, x): + x = self.rescaler(x) + x = self.decoder(x) + return x + + +class Resize(nn.Module): + def __init__(self, in_channels=None, learned=False, mode="bilinear"): + super().__init__() + self.with_conv = learned + self.mode = mode + if self.with_conv: + print(f"Note: {self.__class__.__name} uses learned downsampling and will ignore the fixed {mode} mode") + raise NotImplementedError() + assert in_channels is not None + # no asymmetric padding in torch conv, must do it ourselves + self.conv = torch.nn.Conv2d(in_channels, + in_channels, + kernel_size=4, + stride=2, + padding=1) + + def forward(self, x, scale_factor=1.0): + if scale_factor==1.0: + return x + else: + x = torch.nn.functional.interpolate(x, mode=self.mode, align_corners=False, scale_factor=scale_factor) + return x + +class FirstStagePostProcessor(nn.Module): + + def __init__(self, ch_mult:list, in_channels, + pretrained_model:nn.Module=None, + reshape=False, + n_channels=None, + dropout=0., + pretrained_config=None): + super().__init__() + if pretrained_config is None: + assert pretrained_model is not None, 'Either "pretrained_model" or "pretrained_config" must not be None' + self.pretrained_model = pretrained_model + else: + assert pretrained_config is not None, 'Either "pretrained_model" or "pretrained_config" must not be None' + self.instantiate_pretrained(pretrained_config) + + self.do_reshape = reshape + + if n_channels is None: + n_channels = self.pretrained_model.encoder.ch + + self.proj_norm = Normalize(in_channels,num_groups=in_channels//2) + self.proj = nn.Conv2d(in_channels,n_channels,kernel_size=3, + stride=1,padding=1) + + blocks = [] + downs = [] + ch_in = n_channels + for m in ch_mult: + blocks.append(ResnetBlock(in_channels=ch_in,out_channels=m*n_channels,dropout=dropout)) + ch_in = m * n_channels + downs.append(Downsample(ch_in, with_conv=False)) + + self.model = nn.ModuleList(blocks) + self.downsampler = nn.ModuleList(downs) + + + def instantiate_pretrained(self, config): + model = instantiate_from_config(config) + self.pretrained_model = model.eval() + # self.pretrained_model.train = False + for param in self.pretrained_model.parameters(): + param.requires_grad = False + + + @torch.no_grad() + def encode_with_pretrained(self,x): + c = self.pretrained_model.encode(x) + if isinstance(c, DiagonalGaussianDistribution): + c = c.mode() + return c + + def forward(self,x): + z_fs = self.encode_with_pretrained(x) + z = self.proj_norm(z_fs) + z = self.proj(z) + z = nonlinearity(z) + + for submodel, downmodel in zip(self.model,self.downsampler): + z = submodel(z,temb=None) + z = downmodel(z) + + if self.do_reshape: + z = rearrange(z,'b c h w -> b (h w) c') + return z + diff --git a/lvdm/modules/networks/openaimodel3d_next.py b/lvdm/modules/networks/openaimodel3d_next.py new file mode 100644 index 0000000000000000000000000000000000000000..49ab27a07a16b8bebc16a2172bba9cb4f9a986bd --- /dev/null +++ b/lvdm/modules/networks/openaimodel3d_next.py @@ -0,0 +1,580 @@ +import math +import random +from abc import abstractmethod +from functools import partial + +import numpy as np +import torch +import torch.nn as nn +import torch.nn.functional as F +from einops import rearrange, repeat + +from lvdm.basics import (avg_pool_nd, conv_nd, linear, normalization, + zero_module) +from lvdm.common import checkpoint +from lvdm.models.utils_diffusion import timestep_embedding +from lvdm.modules.attention import SpatialTransformer, TemporalTransformer + + +class TimestepBlock(nn.Module): + """ + Any module where forward() takes timestep embeddings as a second argument. + """ + @abstractmethod + def forward(self, x, emb): + """ + Apply the module to `x` given `emb` timestep embeddings. + """ + + +class TimestepEmbedSequential(nn.Sequential, TimestepBlock): + """ + A sequential module that passes timestep embeddings to the children that + support it as an extra input. + """ + + def forward(self, x, emb, context=None, batch_size=None, is_imgbatch=False): + for layer in self: + if isinstance(layer, TimestepBlock): + x = layer(x, emb, batch_size, is_imgbatch=is_imgbatch) + elif isinstance(layer, SpatialTransformer): + x = layer(x, context) + elif isinstance(layer, TemporalTransformer): + x = rearrange(x, '(b f) c h w -> b c f h w', b=batch_size) + x = layer(x, context, is_imgbatch=is_imgbatch) + x = rearrange(x, 'b c f h w -> (b f) c h w') + else: + x = layer(x,) + return x + + +class Downsample(nn.Module): + """ + A downsampling layer with an optional convolution. + :param channels: channels in the inputs and outputs. + :param use_conv: a bool determining if a convolution is applied. + :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then + downsampling occurs in the inner-two dimensions. + """ + + def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1): + super().__init__() + self.channels = channels + self.out_channels = out_channels or channels + self.use_conv = use_conv + self.dims = dims + stride = 2 if dims != 3 else (1, 2, 2) + if use_conv: + self.op = conv_nd( + dims, self.channels, self.out_channels, 3, stride=stride, padding=padding + ) + else: + assert self.channels == self.out_channels + self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride) + + def forward(self, x): + assert x.shape[1] == self.channels + return self.op(x) + + +class Upsample(nn.Module): + """ + An upsampling layer with an optional convolution. + :param channels: channels in the inputs and outputs. + :param use_conv: a bool determining if a convolution is applied. + :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then + upsampling occurs in the inner-two dimensions. + """ + + def __init__(self, channels, use_conv, dims=2, out_channels=None, padding=1): + super().__init__() + self.channels = channels + self.out_channels = out_channels or channels + self.use_conv = use_conv + self.dims = dims + if use_conv: + self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=padding) + + def forward(self, x): + assert x.shape[1] == self.channels + if self.dims == 3: + x = F.interpolate(x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode='nearest') + else: + x = F.interpolate(x, scale_factor=2, mode='nearest') + if self.use_conv: + x = self.conv(x) + return x + + +class ResBlock(TimestepBlock): + """ + A residual block that can optionally change the number of channels. + :param channels: the number of input channels. + :param emb_channels: the number of timestep embedding channels. + :param dropout: the rate of dropout. + :param out_channels: if specified, the number of out channels. + :param use_conv: if True and out_channels is specified, use a spatial + convolution instead of a smaller 1x1 convolution to change the + channels in the skip connection. + :param dims: determines if the signal is 1D, 2D, or 3D. + :param up: if True, use this block for upsampling. + :param down: if True, use this block for downsampling. + :param use_temporal_conv: if True, use the temporal convolution. + :param use_image_dataset: if True, the temporal parameters will not be optimized. + """ + + def __init__( + self, + channels, + emb_channels, + dropout, + out_channels=None, + use_scale_shift_norm=False, + dims=2, + use_checkpoint=False, + use_conv=False, + up=False, + down=False, + use_temporal_conv=False, + tempspatial_aware=False, + use_image_dataset=False, + ): + super().__init__() + self.channels = channels + self.emb_channels = emb_channels + self.dropout = dropout + self.out_channels = out_channels or channels + self.use_conv = use_conv + self.use_checkpoint = use_checkpoint + self.use_scale_shift_norm = use_scale_shift_norm + self.use_temporal_conv = use_temporal_conv + + self.in_layers = nn.Sequential( + normalization(channels), + nn.SiLU(), + conv_nd(dims, channels, self.out_channels, 3, padding=1), + ) + + self.updown = up or down + + if up: + self.h_upd = Upsample(channels, False, dims) + self.x_upd = Upsample(channels, False, dims) + elif down: + self.h_upd = Downsample(channels, False, dims) + self.x_upd = Downsample(channels, False, dims) + else: + self.h_upd = self.x_upd = nn.Identity() + + self.emb_layers = nn.Sequential( + nn.SiLU(), + nn.Linear( + emb_channels, + 2 * self.out_channels if use_scale_shift_norm else self.out_channels, + ), + ) + self.out_layers = nn.Sequential( + normalization(self.out_channels), + nn.SiLU(), + nn.Dropout(p=dropout), + zero_module(nn.Conv2d(self.out_channels, self.out_channels, 3, padding=1)), + ) + + if self.out_channels == channels: + self.skip_connection = nn.Identity() + elif use_conv: + self.skip_connection = conv_nd(dims, channels, self.out_channels, 3, padding=1) + else: + self.skip_connection = conv_nd(dims, channels, self.out_channels, 1) + + if self.use_temporal_conv: + self.temopral_conv = TemporalConvBlock( + self.out_channels, + self.out_channels, + dropout=0.1, + spatial_aware=tempspatial_aware, + use_image_dataset=use_image_dataset + ) + + def forward(self, x, emb, batch_size=None, is_imgbatch=False): + """ + Apply the block to a Tensor, conditioned on a timestep embedding. + :param x: an [N x C x ...] Tensor of features. + :param emb: an [N x emb_channels] Tensor of timestep embeddings. + :return: an [N x C x ...] Tensor of outputs. + """ + input_tuple = (x, emb) + if self.use_temporal_conv: + forward_tempconv = partial(self._forward, batch_size=batch_size, is_imgbatch=is_imgbatch) + return checkpoint(forward_tempconv, input_tuple, self.parameters(), self.use_checkpoint) + return checkpoint(self._forward, input_tuple, self.parameters(), self.use_checkpoint) + + def _forward(self, x, emb, batch_size=None, is_imgbatch=False): + if self.updown: + in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1] + h = in_rest(x) + h = self.h_upd(h) + x = self.x_upd(x) + h = in_conv(h) + else: + h = self.in_layers(x) + emb_out = self.emb_layers(emb).type(h.dtype) + while len(emb_out.shape) < len(h.shape): + emb_out = emb_out[..., None] + if self.use_scale_shift_norm: + out_norm, out_rest = self.out_layers[0], self.out_layers[1:] + scale, shift = torch.chunk(emb_out, 2, dim=1) + h = out_norm(h) * (1 + scale) + shift + h = out_rest(h) + else: + h = h + emb_out + h = self.out_layers(h) + h = self.skip_connection(x) + h + + if self.use_temporal_conv and batch_size and not is_imgbatch: + h = rearrange(h, '(b t) c h w -> b c t h w', b=batch_size) + h = self.temopral_conv(h) + h = rearrange(h, 'b c t h w -> (b t) c h w') + return h + + +class TemporalConvBlock(nn.Module): + def __init__(self, in_channels, out_channels=None, dropout=0.0, spatial_aware=False, use_image_dataset=False): + super(TemporalConvBlock, self).__init__() + if out_channels is None: + out_channels = in_channels # int(1.5*in_channels) + self.in_channels = in_channels + self.out_channels = out_channels + self.use_image_dataset = use_image_dataset + kernel_shape = (3, 1, 1) if not spatial_aware else (3, 3, 3) + padding_shape = (1, 0, 0) if not spatial_aware else (1, 1, 1) + + # conv layers + self.conv1 = nn.Sequential( + nn.GroupNorm(32, in_channels), nn.SiLU(), + nn.Conv3d(in_channels, out_channels, kernel_shape, padding=padding_shape)) + self.conv2 = nn.Sequential( + nn.GroupNorm(32, out_channels), nn.SiLU(), nn.Dropout(dropout), + nn.Conv3d(out_channels, in_channels, kernel_shape, padding=padding_shape)) + self.conv3 = nn.Sequential( + nn.GroupNorm(32, out_channels), nn.SiLU(), nn.Dropout(dropout), + nn.Conv3d(out_channels, in_channels, (3, 1, 1), padding=(1, 0, 0))) + self.conv4 = nn.Sequential( + nn.GroupNorm(32, out_channels), nn.SiLU(), nn.Dropout(dropout), + nn.Conv3d(out_channels, in_channels, (3, 1, 1), padding=(1, 0, 0))) + + # zero out the last layer params,so the conv block is identity + nn.init.zeros_(self.conv4[-1].weight) + nn.init.zeros_(self.conv4[-1].bias) + + def forward(self, x): + identity = x + x = self.conv1(x) + x = self.conv2(x) + x = self.conv3(x) + x = self.conv4(x) + + if self.use_image_dataset: + x = identity + 0.0 * x + else: + x = identity + x + return x + + +class UNetModel(nn.Module): + """ + The full UNet model with attention and timestep embedding. + :param in_channels: in_channels in the input Tensor. + :param model_channels: base channel count for the model. + :param out_channels: channels in the output Tensor. + :param num_res_blocks: number of residual blocks per downsample. + :param attention_resolutions: a collection of downsample rates at which + attention will take place. May be a set, list, or tuple. + For example, if this contains 4, then at 4x downsampling, attention + will be used. + :param dropout: the dropout probability. + :param channel_mult: channel multiplier for each level of the UNet. + :param conv_resample: if True, use learned convolutions for upsampling and + downsampling. + :param dims: determines if the signal is 1D, 2D, or 3D. + :param num_classes: if specified (as an int), then this model will be + class-conditional with `num_classes` classes. + :param use_checkpoint: use gradient checkpointing to reduce memory usage. + :param num_heads: the number of attention heads in each attention layer. + :param num_heads_channels: if specified, ignore num_heads and instead use + a fixed channel width per attention head. + :param num_heads_upsample: works with num_heads to set a different number + of heads for upsampling. Deprecated. + :param use_scale_shift_norm: use a FiLM-like conditioning mechanism. + :param resblock_updown: use residual blocks for up/downsampling. + :param use_new_attention_order: use a different attention pattern for potentially + increased efficiency. + """ + + def __init__(self, + in_channels, + model_channels, + out_channels, + num_res_blocks, + attention_resolutions, + dropout=0.0, + channel_mult=(1, 2, 4, 8), + conv_resample=True, + dims=2, + context_dim=None, + use_scale_shift_norm=False, + resblock_updown=False, + num_heads=-1, + num_head_channels=-1, + transformer_depth=1, + use_linear=False, + use_checkpoint=False, + temporal_conv=False, + tempspatial_aware=False, + temporal_attention=True, + addition_attention=False, + temporal_selfatt_only=True, + use_relative_position=True, + use_causal_attention=False, + temporal_length=None, + use_image_dataset=False, + use_fp16=False, + micro_condition=False, + temporal_transformer_depth=1 + ): + super(UNetModel, self).__init__() + if num_heads == -1: + assert num_head_channels != -1, 'Either num_heads or num_head_channels has to be set' + if num_head_channels == -1: + assert num_heads != -1, 'Either num_heads or num_head_channels has to be set' + + self.in_channels = in_channels + self.model_channels = model_channels + self.out_channels = out_channels + self.num_res_blocks = num_res_blocks + self.attention_resolutions = attention_resolutions + self.dropout = dropout + self.channel_mult = channel_mult + self.conv_resample = conv_resample + self.temporal_attention = temporal_attention + time_embed_dim = model_channels * 4 + self.use_checkpoint = use_checkpoint + self.dtype = torch.float16 if use_fp16 else torch.float32 + #temporal_selfatt_only = True + self.addition_attention=addition_attention + + + self.time_embed = nn.Sequential( + linear(model_channels, time_embed_dim), + nn.SiLU(), + linear(time_embed_dim, time_embed_dim), + ) + if micro_condition: + self.micro_embed = nn.Sequential( + linear(model_channels, time_embed_dim), + nn.SiLU(), + linear(time_embed_dim, time_embed_dim), + ) + self.micro_condition = micro_condition + + self.input_blocks = nn.ModuleList( + [ + TimestepEmbedSequential(conv_nd(dims, in_channels, model_channels, 3, padding=1)) + ] + ) + if self.addition_attention: + self.init_attn=TimestepEmbedSequential( + TemporalTransformer( + model_channels, + n_heads=8, + d_head=num_head_channels, + depth=transformer_depth, + context_dim=context_dim, + use_checkpoint=use_checkpoint, only_self_att=temporal_selfatt_only, + causal_attention=use_causal_attention, relative_position=use_relative_position, + temporal_length=temporal_length, use_image_dataset=use_image_dataset)) + + input_block_chans = [model_channels] + ch = model_channels + ds = 1 + for level, mult in enumerate(channel_mult): + for _ in range(num_res_blocks): + layers = [ + ResBlock(ch, time_embed_dim, dropout, + out_channels=mult * model_channels, dims=dims, use_checkpoint=use_checkpoint, + use_scale_shift_norm=use_scale_shift_norm, tempspatial_aware=tempspatial_aware, + use_temporal_conv=temporal_conv, use_image_dataset=use_image_dataset + ) + ] + ch = mult * model_channels + if ds in attention_resolutions: + if num_head_channels == -1: + dim_head = ch // num_heads + else: + num_heads = ch // num_head_channels + dim_head = num_head_channels + layers.append( + SpatialTransformer(ch, num_heads, dim_head, + depth=transformer_depth, context_dim=context_dim, use_linear=use_linear, + use_checkpoint=use_checkpoint, disable_self_attn=False + ) + ) + if self.temporal_attention: + layers.append( + TemporalTransformer(ch, num_heads, dim_head, + depth=temporal_transformer_depth, context_dim=context_dim, use_linear=use_linear, + use_checkpoint=use_checkpoint, only_self_att=temporal_selfatt_only, + causal_attention=use_causal_attention, relative_position=use_relative_position, + temporal_length=temporal_length, use_image_dataset=use_image_dataset + ) + ) + self.input_blocks.append(TimestepEmbedSequential(*layers)) + input_block_chans.append(ch) + if level != len(channel_mult) - 1: + out_ch = ch + self.input_blocks.append( + TimestepEmbedSequential( + ResBlock(ch, time_embed_dim, dropout, + out_channels=out_ch, dims=dims, use_checkpoint=use_checkpoint, + use_scale_shift_norm=use_scale_shift_norm, + down=True + ) + if resblock_updown + else Downsample(ch, conv_resample, dims=dims, out_channels=out_ch) + ) + ) + ch = out_ch + input_block_chans.append(ch) + ds *= 2 + + if num_head_channels == -1: + dim_head = ch // num_heads + else: + num_heads = ch // num_head_channels + dim_head = num_head_channels + layers = [ + ResBlock(ch, time_embed_dim, dropout, + dims=dims, use_checkpoint=use_checkpoint, + use_scale_shift_norm=use_scale_shift_norm, tempspatial_aware=tempspatial_aware, + use_temporal_conv=temporal_conv, use_image_dataset=use_image_dataset + ), + SpatialTransformer(ch, num_heads, dim_head, + depth=transformer_depth, context_dim=context_dim, use_linear=use_linear, + use_checkpoint=use_checkpoint, disable_self_attn=False + ) + ] + if self.temporal_attention: + layers.append( + TemporalTransformer(ch, num_heads, dim_head, + depth=temporal_transformer_depth, context_dim=context_dim, use_linear=use_linear, + use_checkpoint=use_checkpoint, only_self_att=temporal_selfatt_only, + causal_attention=use_causal_attention, relative_position=use_relative_position, + temporal_length=temporal_length, use_image_dataset=use_image_dataset + ) + ) + layers.append( + ResBlock(ch, time_embed_dim, dropout, + dims=dims, use_checkpoint=use_checkpoint, + use_scale_shift_norm=use_scale_shift_norm, tempspatial_aware=tempspatial_aware, + use_temporal_conv=temporal_conv, use_image_dataset=use_image_dataset + ) + ) + self.middle_block = TimestepEmbedSequential(*layers) + + self.output_blocks = nn.ModuleList([]) + for level, mult in list(enumerate(channel_mult))[::-1]: + for i in range(num_res_blocks + 1): + ich = input_block_chans.pop() + layers = [ + ResBlock(ch + ich, time_embed_dim, dropout, + out_channels=mult * model_channels, dims=dims, use_checkpoint=use_checkpoint, + use_scale_shift_norm=use_scale_shift_norm, tempspatial_aware=tempspatial_aware, + use_temporal_conv=temporal_conv, use_image_dataset=use_image_dataset + ) + ] + ch = model_channels * mult + if ds in attention_resolutions: + if num_head_channels == -1: + dim_head = ch // num_heads + else: + num_heads = ch // num_head_channels + dim_head = num_head_channels + layers.append( + SpatialTransformer(ch, num_heads, dim_head, + depth=transformer_depth, context_dim=context_dim, use_linear=use_linear, + use_checkpoint=use_checkpoint, disable_self_attn=False + ) + ) + if self.temporal_attention: + layers.append( + TemporalTransformer(ch, num_heads, dim_head, + depth=temporal_transformer_depth, context_dim=context_dim, use_linear=use_linear, + use_checkpoint=use_checkpoint, only_self_att=temporal_selfatt_only, + causal_attention=use_causal_attention, relative_position=use_relative_position, + temporal_length=temporal_length, use_image_dataset=use_image_dataset + ) + ) + if level and i == num_res_blocks: + out_ch = ch + layers.append( + ResBlock(ch, time_embed_dim, dropout, + out_channels=out_ch, dims=dims, use_checkpoint=use_checkpoint, + use_scale_shift_norm=use_scale_shift_norm, + up=True + ) + if resblock_updown + else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch) + ) + ds //= 2 + self.output_blocks.append(TimestepEmbedSequential(*layers)) + + self.out = nn.Sequential( + normalization(ch), + nn.SiLU(), + zero_module(conv_nd(dims, model_channels, out_channels, 3, padding=1)), + ) + + def forward(self, x, timesteps, context=None, y=None, features_adapter=None, is_imgbatch=False, **kwargs): + b,_,t,_,_ = x.shape + + t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False) + emb = self.time_embed(t_emb) + if self.micro_condition and y is not None: + micro_emb = timestep_embedding(y, self.model_channels, repeat_only=False) + emb = emb + self.micro_embed(micro_emb) + + ## repeat t times for context [(b t) 77 768] & time embedding + if not is_imgbatch: + context = context.repeat_interleave(repeats=t, dim=0) + emb = emb.repeat_interleave(repeats=t, dim=0) + + ## always in shape (b t) c h w, except for temporal layer + x = rearrange(x, 'b c t h w -> (b t) c h w') + if features_adapter is not None: + features_adapter = [rearrange(feature, 'b c t h w -> (b t) c h w') for feature in features_adapter] + + h = x.type(self.dtype) + adapter_idx = 0 + hs = [] + for id, module in enumerate(self.input_blocks): + h = module(h, emb, context=context, batch_size=b,is_imgbatch=is_imgbatch) + if id ==0 and self.addition_attention: + h = self.init_attn(h, emb, context=context, batch_size=b,is_imgbatch=is_imgbatch) + ## plug-in adapter features + if ((id+1)%3 == 0) and features_adapter is not None: + h = h + features_adapter[adapter_idx] + adapter_idx += 1 + hs.append(h) + if features_adapter is not None: + assert len(features_adapter)==adapter_idx, 'Wrong features_adapter' + + h = self.middle_block(h, emb, context=context, batch_size=b, is_imgbatch=is_imgbatch) + for module in self.output_blocks: + h = torch.cat([h, hs.pop()], dim=1) + h = module(h, emb, context=context, batch_size=b, is_imgbatch=is_imgbatch) + h = h.type(x.dtype) + y = self.out(h) + + # reshape back to (b c t h w) + y = rearrange(y, '(b t) c h w -> b c t h w', b=b) + return y diff --git a/lvdm/modules/x_transformer.py b/lvdm/modules/x_transformer.py new file mode 100644 index 0000000000000000000000000000000000000000..396344a7d4f353c523dac43e74f2a50efbf68b8e --- /dev/null +++ b/lvdm/modules/x_transformer.py @@ -0,0 +1,641 @@ +"""shout-out to https://github.com/lucidrains/x-transformers/tree/main/x_transformers""" +from functools import partial +from inspect import isfunction +from collections import namedtuple +from einops import rearrange, repeat + +import torch +from torch import nn, einsum +import torch.nn.functional as F + +# constants +DEFAULT_DIM_HEAD = 64 + +Intermediates = namedtuple('Intermediates', [ + 'pre_softmax_attn', + 'post_softmax_attn' +]) + +LayerIntermediates = namedtuple('Intermediates', [ + 'hiddens', + 'attn_intermediates' +]) + + +class AbsolutePositionalEmbedding(nn.Module): + def __init__(self, dim, max_seq_len): + super().__init__() + self.emb = nn.Embedding(max_seq_len, dim) + self.init_() + + def init_(self): + nn.init.normal_(self.emb.weight, std=0.02) + + def forward(self, x): + n = torch.arange(x.shape[1], device=x.device) + return self.emb(n)[None, :, :] + + +class FixedPositionalEmbedding(nn.Module): + def __init__(self, dim): + super().__init__() + inv_freq = 1. / (10000 ** (torch.arange(0, dim, 2).float() / dim)) + self.register_buffer('inv_freq', inv_freq) + + def forward(self, x, seq_dim=1, offset=0): + t = torch.arange(x.shape[seq_dim], device=x.device).type_as(self.inv_freq) + offset + sinusoid_inp = torch.einsum('i , j -> i j', t, self.inv_freq) + emb = torch.cat((sinusoid_inp.sin(), sinusoid_inp.cos()), dim=-1) + return emb[None, :, :] + + +# helpers + +def exists(val): + return val is not None + + +def default(val, d): + if exists(val): + return val + return d() if isfunction(d) else d + + +def always(val): + def inner(*args, **kwargs): + return val + return inner + + +def not_equals(val): + def inner(x): + return x != val + return inner + + +def equals(val): + def inner(x): + return x == val + return inner + + +def max_neg_value(tensor): + return -torch.finfo(tensor.dtype).max + + +# keyword argument helpers + +def pick_and_pop(keys, d): + values = list(map(lambda key: d.pop(key), keys)) + return dict(zip(keys, values)) + + +def group_dict_by_key(cond, d): + return_val = [dict(), dict()] + for key in d.keys(): + match = bool(cond(key)) + ind = int(not match) + return_val[ind][key] = d[key] + return (*return_val,) + + +def string_begins_with(prefix, str): + return str.startswith(prefix) + + +def group_by_key_prefix(prefix, d): + return group_dict_by_key(partial(string_begins_with, prefix), d) + + +def groupby_prefix_and_trim(prefix, d): + kwargs_with_prefix, kwargs = group_dict_by_key(partial(string_begins_with, prefix), d) + kwargs_without_prefix = dict(map(lambda x: (x[0][len(prefix):], x[1]), tuple(kwargs_with_prefix.items()))) + return kwargs_without_prefix, kwargs + + +# classes +class Scale(nn.Module): + def __init__(self, value, fn): + super().__init__() + self.value = value + self.fn = fn + + def forward(self, x, **kwargs): + x, *rest = self.fn(x, **kwargs) + return (x * self.value, *rest) + + +class Rezero(nn.Module): + def __init__(self, fn): + super().__init__() + self.fn = fn + self.g = nn.Parameter(torch.zeros(1)) + + def forward(self, x, **kwargs): + x, *rest = self.fn(x, **kwargs) + return (x * self.g, *rest) + + +class ScaleNorm(nn.Module): + def __init__(self, dim, eps=1e-5): + super().__init__() + self.scale = dim ** -0.5 + self.eps = eps + self.g = nn.Parameter(torch.ones(1)) + + def forward(self, x): + norm = torch.norm(x, dim=-1, keepdim=True) * self.scale + return x / norm.clamp(min=self.eps) * self.g + + +class RMSNorm(nn.Module): + def __init__(self, dim, eps=1e-8): + super().__init__() + self.scale = dim ** -0.5 + self.eps = eps + self.g = nn.Parameter(torch.ones(dim)) + + def forward(self, x): + norm = torch.norm(x, dim=-1, keepdim=True) * self.scale + return x / norm.clamp(min=self.eps) * self.g + + +class Residual(nn.Module): + def forward(self, x, residual): + return x + residual + + +class GRUGating(nn.Module): + def __init__(self, dim): + super().__init__() + self.gru = nn.GRUCell(dim, dim) + + def forward(self, x, residual): + gated_output = self.gru( + rearrange(x, 'b n d -> (b n) d'), + rearrange(residual, 'b n d -> (b n) d') + ) + + return gated_output.reshape_as(x) + + +# feedforward + +class GEGLU(nn.Module): + def __init__(self, dim_in, dim_out): + super().__init__() + self.proj = nn.Linear(dim_in, dim_out * 2) + + def forward(self, x): + x, gate = self.proj(x).chunk(2, dim=-1) + return x * F.gelu(gate) + + +class FeedForward(nn.Module): + def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.): + super().__init__() + inner_dim = int(dim * mult) + dim_out = default(dim_out, dim) + project_in = nn.Sequential( + nn.Linear(dim, inner_dim), + nn.GELU() + ) if not glu else GEGLU(dim, inner_dim) + + self.net = nn.Sequential( + project_in, + nn.Dropout(dropout), + nn.Linear(inner_dim, dim_out) + ) + + def forward(self, x): + return self.net(x) + + +# attention. +class Attention(nn.Module): + def __init__( + self, + dim, + dim_head=DEFAULT_DIM_HEAD, + heads=8, + causal=False, + mask=None, + talking_heads=False, + sparse_topk=None, + use_entmax15=False, + num_mem_kv=0, + dropout=0., + on_attn=False + ): + super().__init__() + if use_entmax15: + raise NotImplementedError("Check out entmax activation instead of softmax activation!") + self.scale = dim_head ** -0.5 + self.heads = heads + self.causal = causal + self.mask = mask + + inner_dim = dim_head * heads + + self.to_q = nn.Linear(dim, inner_dim, bias=False) + self.to_k = nn.Linear(dim, inner_dim, bias=False) + self.to_v = nn.Linear(dim, inner_dim, bias=False) + self.dropout = nn.Dropout(dropout) + + # talking heads + self.talking_heads = talking_heads + if talking_heads: + self.pre_softmax_proj = nn.Parameter(torch.randn(heads, heads)) + self.post_softmax_proj = nn.Parameter(torch.randn(heads, heads)) + + # explicit topk sparse attention + self.sparse_topk = sparse_topk + + # entmax + #self.attn_fn = entmax15 if use_entmax15 else F.softmax + self.attn_fn = F.softmax + + # add memory key / values + self.num_mem_kv = num_mem_kv + if num_mem_kv > 0: + self.mem_k = nn.Parameter(torch.randn(heads, num_mem_kv, dim_head)) + self.mem_v = nn.Parameter(torch.randn(heads, num_mem_kv, dim_head)) + + # attention on attention + self.attn_on_attn = on_attn + self.to_out = nn.Sequential(nn.Linear(inner_dim, dim * 2), nn.GLU()) if on_attn else nn.Linear(inner_dim, dim) + + def forward( + self, + x, + context=None, + mask=None, + context_mask=None, + rel_pos=None, + sinusoidal_emb=None, + prev_attn=None, + mem=None + ): + b, n, _, h, talking_heads, device = *x.shape, self.heads, self.talking_heads, x.device + kv_input = default(context, x) + + q_input = x + k_input = kv_input + v_input = kv_input + + if exists(mem): + k_input = torch.cat((mem, k_input), dim=-2) + v_input = torch.cat((mem, v_input), dim=-2) + + if exists(sinusoidal_emb): + # in shortformer, the query would start at a position offset depending on the past cached memory + offset = k_input.shape[-2] - q_input.shape[-2] + q_input = q_input + sinusoidal_emb(q_input, offset=offset) + k_input = k_input + sinusoidal_emb(k_input) + + q = self.to_q(q_input) + k = self.to_k(k_input) + v = self.to_v(v_input) + + q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h=h), (q, k, v)) + + input_mask = None + if any(map(exists, (mask, context_mask))): + q_mask = default(mask, lambda: torch.ones((b, n), device=device).bool()) + k_mask = q_mask if not exists(context) else context_mask + k_mask = default(k_mask, lambda: torch.ones((b, k.shape[-2]), device=device).bool()) + q_mask = rearrange(q_mask, 'b i -> b () i ()') + k_mask = rearrange(k_mask, 'b j -> b () () j') + input_mask = q_mask * k_mask + + if self.num_mem_kv > 0: + mem_k, mem_v = map(lambda t: repeat(t, 'h n d -> b h n d', b=b), (self.mem_k, self.mem_v)) + k = torch.cat((mem_k, k), dim=-2) + v = torch.cat((mem_v, v), dim=-2) + if exists(input_mask): + input_mask = F.pad(input_mask, (self.num_mem_kv, 0), value=True) + + dots = einsum('b h i d, b h j d -> b h i j', q, k) * self.scale + mask_value = max_neg_value(dots) + + if exists(prev_attn): + dots = dots + prev_attn + + pre_softmax_attn = dots + + if talking_heads: + dots = einsum('b h i j, h k -> b k i j', dots, self.pre_softmax_proj).contiguous() + + if exists(rel_pos): + dots = rel_pos(dots) + + if exists(input_mask): + dots.masked_fill_(~input_mask, mask_value) + del input_mask + + if self.causal: + i, j = dots.shape[-2:] + r = torch.arange(i, device=device) + mask = rearrange(r, 'i -> () () i ()') < rearrange(r, 'j -> () () () j') + mask = F.pad(mask, (j - i, 0), value=False) + dots.masked_fill_(mask, mask_value) + del mask + + if exists(self.sparse_topk) and self.sparse_topk < dots.shape[-1]: + top, _ = dots.topk(self.sparse_topk, dim=-1) + vk = top[..., -1].unsqueeze(-1).expand_as(dots) + mask = dots < vk + dots.masked_fill_(mask, mask_value) + del mask + + attn = self.attn_fn(dots, dim=-1) + post_softmax_attn = attn + + attn = self.dropout(attn) + + if talking_heads: + attn = einsum('b h i j, h k -> b k i j', attn, self.post_softmax_proj).contiguous() + + out = einsum('b h i j, b h j d -> b h i d', attn, v) + out = rearrange(out, 'b h n d -> b n (h d)') + + intermediates = Intermediates( + pre_softmax_attn=pre_softmax_attn, + post_softmax_attn=post_softmax_attn + ) + + return self.to_out(out), intermediates + + +class AttentionLayers(nn.Module): + def __init__( + self, + dim, + depth, + heads=8, + causal=False, + cross_attend=False, + only_cross=False, + use_scalenorm=False, + use_rmsnorm=False, + use_rezero=False, + rel_pos_num_buckets=32, + rel_pos_max_distance=128, + position_infused_attn=False, + custom_layers=None, + sandwich_coef=None, + par_ratio=None, + residual_attn=False, + cross_residual_attn=False, + macaron=False, + pre_norm=True, + gate_residual=False, + **kwargs + ): + super().__init__() + ff_kwargs, kwargs = groupby_prefix_and_trim('ff_', kwargs) + attn_kwargs, _ = groupby_prefix_and_trim('attn_', kwargs) + + dim_head = attn_kwargs.get('dim_head', DEFAULT_DIM_HEAD) + + self.dim = dim + self.depth = depth + self.layers = nn.ModuleList([]) + + self.has_pos_emb = position_infused_attn + self.pia_pos_emb = FixedPositionalEmbedding(dim) if position_infused_attn else None + self.rotary_pos_emb = always(None) + + assert rel_pos_num_buckets <= rel_pos_max_distance, 'number of relative position buckets must be less than the relative position max distance' + self.rel_pos = None + + self.pre_norm = pre_norm + + self.residual_attn = residual_attn + self.cross_residual_attn = cross_residual_attn + + norm_class = ScaleNorm if use_scalenorm else nn.LayerNorm + norm_class = RMSNorm if use_rmsnorm else norm_class + norm_fn = partial(norm_class, dim) + + norm_fn = nn.Identity if use_rezero else norm_fn + branch_fn = Rezero if use_rezero else None + + if cross_attend and not only_cross: + default_block = ('a', 'c', 'f') + elif cross_attend and only_cross: + default_block = ('c', 'f') + else: + default_block = ('a', 'f') + + if macaron: + default_block = ('f',) + default_block + + if exists(custom_layers): + layer_types = custom_layers + elif exists(par_ratio): + par_depth = depth * len(default_block) + assert 1 < par_ratio <= par_depth, 'par ratio out of range' + default_block = tuple(filter(not_equals('f'), default_block)) + par_attn = par_depth // par_ratio + depth_cut = par_depth * 2 // 3 # 2 / 3 attention layer cutoff suggested by PAR paper + par_width = (depth_cut + depth_cut // par_attn) // par_attn + assert len(default_block) <= par_width, 'default block is too large for par_ratio' + par_block = default_block + ('f',) * (par_width - len(default_block)) + par_head = par_block * par_attn + layer_types = par_head + ('f',) * (par_depth - len(par_head)) + elif exists(sandwich_coef): + assert sandwich_coef > 0 and sandwich_coef <= depth, 'sandwich coefficient should be less than the depth' + layer_types = ('a',) * sandwich_coef + default_block * (depth - sandwich_coef) + ('f',) * sandwich_coef + else: + layer_types = default_block * depth + + self.layer_types = layer_types + self.num_attn_layers = len(list(filter(equals('a'), layer_types))) + + for layer_type in self.layer_types: + if layer_type == 'a': + layer = Attention(dim, heads=heads, causal=causal, **attn_kwargs) + elif layer_type == 'c': + layer = Attention(dim, heads=heads, **attn_kwargs) + elif layer_type == 'f': + layer = FeedForward(dim, **ff_kwargs) + layer = layer if not macaron else Scale(0.5, layer) + else: + raise Exception(f'invalid layer type {layer_type}') + + if isinstance(layer, Attention) and exists(branch_fn): + layer = branch_fn(layer) + + if gate_residual: + residual_fn = GRUGating(dim) + else: + residual_fn = Residual() + + self.layers.append(nn.ModuleList([ + norm_fn(), + layer, + residual_fn + ])) + + def forward( + self, + x, + context=None, + mask=None, + context_mask=None, + mems=None, + return_hiddens=False + ): + hiddens = [] + intermediates = [] + prev_attn = None + prev_cross_attn = None + + mems = mems.copy() if exists(mems) else [None] * self.num_attn_layers + + for ind, (layer_type, (norm, block, residual_fn)) in enumerate(zip(self.layer_types, self.layers)): + is_last = ind == (len(self.layers) - 1) + + if layer_type == 'a': + hiddens.append(x) + layer_mem = mems.pop(0) + + residual = x + + if self.pre_norm: + x = norm(x) + + if layer_type == 'a': + out, inter = block(x, mask=mask, sinusoidal_emb=self.pia_pos_emb, rel_pos=self.rel_pos, + prev_attn=prev_attn, mem=layer_mem) + elif layer_type == 'c': + out, inter = block(x, context=context, mask=mask, context_mask=context_mask, prev_attn=prev_cross_attn) + elif layer_type == 'f': + out = block(x) + + x = residual_fn(out, residual) + + if layer_type in ('a', 'c'): + intermediates.append(inter) + + if layer_type == 'a' and self.residual_attn: + prev_attn = inter.pre_softmax_attn + elif layer_type == 'c' and self.cross_residual_attn: + prev_cross_attn = inter.pre_softmax_attn + + if not self.pre_norm and not is_last: + x = norm(x) + + if return_hiddens: + intermediates = LayerIntermediates( + hiddens=hiddens, + attn_intermediates=intermediates + ) + + return x, intermediates + + return x + + +class Encoder(AttentionLayers): + def __init__(self, **kwargs): + assert 'causal' not in kwargs, 'cannot set causality on encoder' + super().__init__(causal=False, **kwargs) + + + +class TransformerWrapper(nn.Module): + def __init__( + self, + *, + num_tokens, + max_seq_len, + attn_layers, + emb_dim=None, + max_mem_len=0., + emb_dropout=0., + num_memory_tokens=None, + tie_embedding=False, + use_pos_emb=True + ): + super().__init__() + assert isinstance(attn_layers, AttentionLayers), 'attention layers must be one of Encoder or Decoder' + + dim = attn_layers.dim + emb_dim = default(emb_dim, dim) + + self.max_seq_len = max_seq_len + self.max_mem_len = max_mem_len + self.num_tokens = num_tokens + + self.token_emb = nn.Embedding(num_tokens, emb_dim) + self.pos_emb = AbsolutePositionalEmbedding(emb_dim, max_seq_len) if ( + use_pos_emb and not attn_layers.has_pos_emb) else always(0) + self.emb_dropout = nn.Dropout(emb_dropout) + + self.project_emb = nn.Linear(emb_dim, dim) if emb_dim != dim else nn.Identity() + self.attn_layers = attn_layers + self.norm = nn.LayerNorm(dim) + + self.init_() + + self.to_logits = nn.Linear(dim, num_tokens) if not tie_embedding else lambda t: t @ self.token_emb.weight.t() + + # memory tokens (like [cls]) from Memory Transformers paper + num_memory_tokens = default(num_memory_tokens, 0) + self.num_memory_tokens = num_memory_tokens + if num_memory_tokens > 0: + self.memory_tokens = nn.Parameter(torch.randn(num_memory_tokens, dim)) + + # let funnel encoder know number of memory tokens, if specified + if hasattr(attn_layers, 'num_memory_tokens'): + attn_layers.num_memory_tokens = num_memory_tokens + + def init_(self): + nn.init.normal_(self.token_emb.weight, std=0.02) + + def forward( + self, + x, + return_embeddings=False, + mask=None, + return_mems=False, + return_attn=False, + mems=None, + **kwargs + ): + b, n, device, num_mem = *x.shape, x.device, self.num_memory_tokens + x = self.token_emb(x) + x += self.pos_emb(x) + x = self.emb_dropout(x) + + x = self.project_emb(x) + + if num_mem > 0: + mem = repeat(self.memory_tokens, 'n d -> b n d', b=b) + x = torch.cat((mem, x), dim=1) + + # auto-handle masking after appending memory tokens + if exists(mask): + mask = F.pad(mask, (num_mem, 0), value=True) + + x, intermediates = self.attn_layers(x, mask=mask, mems=mems, return_hiddens=True, **kwargs) + x = self.norm(x) + + mem, x = x[:, :num_mem], x[:, num_mem:] + + out = self.to_logits(x) if not return_embeddings else x + + if return_mems: + hiddens = intermediates.hiddens + new_mems = list(map(lambda pair: torch.cat(pair, dim=-2), zip(mems, hiddens))) if exists(mems) else hiddens + new_mems = list(map(lambda t: t[..., -self.max_mem_len:, :].detach(), new_mems)) + return out, new_mems + + if return_attn: + attn_maps = list(map(lambda t: t.post_softmax_attn, intermediates.attn_intermediates)) + return out, attn_maps + + return out + diff --git a/main/evaluation/motionctrl_inference.py b/main/evaluation/motionctrl_inference.py new file mode 100644 index 0000000000000000000000000000000000000000..b37e4b0b585a13a5dfd5151515114c5110ea6854 --- /dev/null +++ b/main/evaluation/motionctrl_inference.py @@ -0,0 +1,353 @@ +import argparse +import datetime +import glob +import json +import math +import os +import sys +import time +from collections import OrderedDict + +import cv2 +import numpy as np +import torch +import torchvision +## note: decord should be imported after torch +from omegaconf import OmegaConf +from pytorch_lightning import seed_everything +from tqdm import tqdm + +sys.path.insert(1, os.path.join(sys.path[0], '..', '..')) +from lvdm.models.samplers.ddim import DDIMSampler +from main.evaluation.motionctrl_prompts_camerapose_trajs import ( + both_prompt_camerapose_traj, cmcm_prompt_camerapose, omom_prompt_traj) +from utils.utils import instantiate_from_config + +DEFAULT_NEGATIVE_PROMPT = 'blur, haze, deformed iris, deformed pupils, semi-realistic, cgi, 3d, render, '\ + 'sketch, cartoon, drawing, anime, mutated hands and fingers, deformed, distorted, '\ + 'disfigured, poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, '\ + 'floating limbs, disconnected limbs, mutation, mutated, ugly, disgusting, amputation' + +post_prompt = 'Ultra-detail, masterpiece, best quality, cinematic lighting, 8k uhd, dslr, soft lighting, film grain, Fujifilm XT3' + + +def load_model_checkpoint(model, ckpt, adapter_ckpt=None): + if adapter_ckpt: + ## main model + state_dict = torch.load(ckpt, map_location="cpu") + if "state_dict" in list(state_dict.keys()): + state_dict = state_dict["state_dict"] + result = model.load_state_dict(state_dict, strict=False) + else: + # deepspeed + new_pl_sd = OrderedDict() + for key in state_dict['module'].keys(): + new_pl_sd[key[16:]]=state_dict['module'][key] + result = model.load_state_dict(new_pl_sd, strict=False) + print(result) + print('>>> model checkpoint loaded.') + ## adapter + state_dict = torch.load(adapter_ckpt, map_location="cpu") + if "state_dict" in list(state_dict.keys()): + state_dict = state_dict["state_dict"] + model.adapter.load_state_dict(state_dict, strict=True) + print('>>> adapter checkpoint loaded.') + else: + state_dict = torch.load(ckpt, map_location="cpu") + if "state_dict" in list(state_dict.keys()): + state_dict = state_dict["state_dict"] + model.load_state_dict(state_dict, strict=False) + else: + # deepspeed + new_pl_sd = OrderedDict() + for key in state_dict['module'].keys(): + new_pl_sd[key[16:]]=state_dict['module'][key] + model.load_state_dict(new_pl_sd) + + print('>>> model checkpoint loaded.') + return model + +def load_trajs(cond_dir, trajs): + traj_files = [f'{cond_dir}/trajectories/{traj}.npy' for traj in trajs] + + data_list = [] + traj_name = [] + + for idx in range(len(traj_files)): + traj_name.append(traj_files[idx].split('/')[-1].split('.')[0]) + data_list.append(torch.tensor(np.load(traj_files[idx])).permute(3, 0, 1, 2).float()) # [t,h,w,c] -> [c,t,h,w] + + return data_list, traj_name + +def load_camera_pose(cond_dir, camera_poses): + + pose_file = [f'{cond_dir}/camera_poses/{pose}.json' for pose in camera_poses] + pose_sample_num = len(pose_file) + + data_list = [] + pose_name = [] + + for idx in range(pose_sample_num): + cur_pose_name = camera_poses[idx].replace('test_camera_', '') + pose_name.append(cur_pose_name) + + with open(pose_file[idx], 'r') as f: + pose = json.load(f) + pose = np.array(pose) # [t, 12] + pose = torch.tensor(pose).float() # [t, 12] + data_list.append(pose) + + return data_list, pose_name + +def save_results(samples, filename, savedir, fps=10): + ## save prompt + + ## save video + videos = [samples] + savedirs = [savedir] + for idx, video in enumerate(videos): + if video is None: + continue + # b,c,t,h,w + video = video.detach().cpu() + video = torch.clamp(video.float(), -1., 1.) + n = video.shape[0] + video = video.permute(2, 0, 1, 3, 4) # t,n,c,h,w + frame_grids = [torchvision.utils.make_grid(framesheet, nrow=int(n)) for framesheet in video] #[3, 1*h, n*w] + grid = torch.stack(frame_grids, dim=0) # stack in temporal dim [t, 3, n*h, w] + grid = (grid + 1.0) / 2.0 + grid = (grid * 255).to(torch.uint8).permute(0, 2, 3, 1) + path = os.path.join(savedirs[idx], "%s.mp4"%filename) + torchvision.io.write_video(path, grid, fps=fps, video_codec='h264', options={'crf': '10'}) + +def motionctrl_sample( + model, + prompts, + noise_shape, + camera_poses=None, + trajs=None, + n_samples=1, + unconditional_guidance_scale=1.0, + unconditional_guidance_scale_temporal=None, + ddim_steps=50, + ddim_eta=1., + **kwargs): + + ddim_sampler = DDIMSampler(model) + batch_size = noise_shape[0] + ## get condition embeddings (support single prompt only) + if isinstance(prompts, str): + prompts = [prompts] + + for i in range(len(prompts)): + prompts[i] = f'{prompts[i]}, {post_prompt}' + + cond = model.get_learned_conditioning(prompts) + if camera_poses is not None: + RT = camera_poses[..., None] + else: + RT = None + + if trajs is not None: + traj_features = model.get_traj_features(trajs) + else: + traj_features = None + + if unconditional_guidance_scale != 1.0: + # prompts = batch_size * [""] + prompts = batch_size * [DEFAULT_NEGATIVE_PROMPT] + uc = model.get_learned_conditioning(prompts) + if traj_features is not None: + un_motion = model.get_traj_features(torch.zeros_like(trajs)) + else: + un_motion = None + uc = {"features_adapter": un_motion, "uc": uc} + else: + uc = None + + batch_variants = [] + for _ in range(n_samples): + if ddim_sampler is not None: + samples, _ = ddim_sampler.sample(S=ddim_steps, + conditioning=cond, + batch_size=noise_shape[0], + shape=noise_shape[1:], + verbose=False, + unconditional_guidance_scale=unconditional_guidance_scale, + unconditional_conditioning=uc, + eta=ddim_eta, + temporal_length=noise_shape[2], + conditional_guidance_scale_temporal=unconditional_guidance_scale_temporal, + features_adapter=traj_features, + pose_emb=RT, + **kwargs + ) + ## reconstruct from latent to pixel space + batch_images = model.decode_first_stage(samples) + batch_variants.append(batch_images) + ## variants, batch, c, t, h, w + batch_variants = torch.stack(batch_variants) + return batch_variants.permute(1, 0, 2, 3, 4, 5) + +def run_inference(args, gpu_num, gpu_no): + ## model config + config = OmegaConf.load(args.base) + model_config = config.pop("model", OmegaConf.create()) + model = instantiate_from_config(model_config) + model = model.cuda(gpu_no) + assert os.path.exists(args.ckpt_path), f"Error: checkpoint {args.ckpt_path} Not Found!" + print(f"Loading checkpoint from {args.ckpt_path}") + model = load_model_checkpoint(model, args.ckpt_path, args.adapter_ckpt) + model.eval() + + ## run over data + assert (args.height % 16 == 0) and (args.width % 16 == 0), "Error: image size [h,w] should be multiples of 16!" + + ## latent noise shape + h, w = args.height // 8, args.width // 8 + channels = model.channels + frames = model.temporal_length + noise_shape = [args.bs, channels, frames, h, w] + + savedir = os.path.join(args.savedir, "samples") + os.makedirs(savedir, exist_ok=True) + + if args.condtype == 'camera_motion': + prompt_list = cmcm_prompt_camerapose['prompts'] + camera_pose_list, pose_name = load_camera_pose(args.cond_dir, cmcm_prompt_camerapose['camera_poses']) + traj_list = None + save_name_list = [] + for i in range(len(pose_name)): + save_name_list.append(f"{pose_name[i]}__{prompt_list[i].replace(' ', '_').replace(',', '')}") + elif args.condtype == 'object_motion': + prompt_list = omom_prompt_traj['prompts'] + traj_list, traj_name = load_trajs(args.cond_dir, omom_prompt_traj['trajs']) + camera_pose_list = None + save_name_list = [] + for i in range(len(traj_name)): + save_name_list.append(f"{traj_name[i]}__{prompt_list[i].replace(' ', '_').replace(',', '')}") + elif args.condtype == 'both': + prompt_list = both_prompt_camerapose_traj['prompts'] + camera_pose_list, pose_name = load_camera_pose(args.cond_dir, both_prompt_camerapose_traj['camera_poses']) + traj_list, traj_name = load_trajs(args.cond_dir, both_prompt_camerapose_traj['trajs']) + save_name_list = [] + for i in range(len(pose_name)): + save_name_list.append(f"{pose_name[i]}__{traj_name[i]}__{prompt_list[i].replace(' ', '_').replace(',', '')}") + + num_samples = len(prompt_list) + samples_split = num_samples // gpu_num + print('Prompts testing [rank:%d] %d/%d samples loaded.'%(gpu_no, samples_split, num_samples)) + #indices = random.choices(list(range(0, num_samples)), k=samples_per_device) + indices = list(range(samples_split*gpu_no, samples_split*(gpu_no+1))) + prompt_list_rank = [prompt_list[i] for i in indices] + camera_pose_list_rank = None if camera_pose_list is None else [camera_pose_list[i] for i in indices] + traj_list_rank = None if traj_list is None else [traj_list[i] for i in indices] + save_name_list_rank = [save_name_list[i] for i in indices] + + start = time.time() + for idx, indice in tqdm(enumerate(range(0, len(prompt_list_rank), args.bs)), desc='Sample Batch'): + prompts = prompt_list_rank[indice:indice+args.bs] + camera_poses = None if camera_pose_list_rank is None else camera_pose_list_rank[indice:indice+args.bs] + trajs = None if traj_list_rank is None else traj_list_rank[indice:indice+args.bs] + save_name = save_name_list_rank[indice:indice+args.bs] + print(f'Processing {save_name}') + + if camera_poses is not None: + camera_poses = torch.stack(camera_poses, dim=0).to("cuda") + if trajs is not None: + trajs = torch.stack(trajs, dim=0).to("cuda") + + batch_samples = motionctrl_sample( + model, + prompts, + noise_shape, + camera_poses=camera_poses, + trajs=trajs, + n_samples=args.n_samples, + unconditional_guidance_scale=args.unconditional_guidance_scale, + unconditional_guidance_scale_temporal=args.unconditional_guidance_scale_temporal, + ddim_steps=args.ddim_steps, + ddim_eta=args.ddim_eta, + cond_T = args.cond_T, + ) + + ## save each example individually + for nn, samples in enumerate(batch_samples): + ## samples : [n_samples,c,t,h,w] + prompt = prompts[nn] + name = save_name[nn] + if len(name) > 90: + name = name[:90] + filename = f'{name}_{idx*args.bs+nn:04d}_randk{gpu_no}' + + save_results(samples, filename, savedir, fps=10) + if args.save_imgs: + parts = save_name[nn].split('__') + if len(parts) == 2: + cond_name = parts[0] + prname = prompts[nn].replace(' ', '_').replace(',', '') + cur_outdir = os.path.join(savedir, cond_name, prname) + elif len(parts) == 3: + poname, trajname, _ = save_name[nn].split('__') + prname = prompts[nn].replace(' ', '_').replace(',', '') + cur_outdir = os.path.join(savedir, poname, trajname, prname) + else: + raise NotImplementedError + os.makedirs(cur_outdir, exist_ok=True) + save_images(samples, cur_outdir) + if nn % 100 == 0: + print(f'Finish {nn}/{len(batch_samples)}') + + print(f"Saved in {args.savedir}. Time used: {(time.time() - start):.2f} seconds") + +def save_images(samples, savedir): + ## samples : [n_samples,c,t,h,w] + n_samples, c, t, h, w = samples.shape + samples = torch.clamp(samples, -1.0, 1.0) + samples = (samples + 1.0) / 2.0 + samples = (samples * 255).detach().cpu().numpy().astype(np.uint8) + for i in range(n_samples): + cur_outdir = os.path.join(savedir, f'{i}/images') + os.makedirs(cur_outdir, exist_ok=True) + + for j in range(t): + img = samples[i,:,j,:,:] + img = np.transpose(img, (1,2,0)) + img = img[:,:,::-1] # BGR to RGB + path = os.path.join(cur_outdir, f'{j:04d}.png') + cv2.imwrite(path, img) + +def get_parser(): + parser = argparse.ArgumentParser() + parser.add_argument("--savedir", type=str, default=None, help="results saving path") + parser.add_argument("--ckpt_path", type=str, default=None, help="checkpoint path") + parser.add_argument("--adapter_ckpt", type=str, default=None, help="adapter checkpoint path") + parser.add_argument("--base", type=str, help="config (yaml) path") + parser.add_argument("--condtype", default='frame', type=str, help="conditon type: {frame, depth, adapter}") + parser.add_argument("--prompt_dir", type=str, default=None, help="a data dir containing videos and prompts") + parser.add_argument("--n_samples", type=int, default=1, help="num of samples per prompt",) + parser.add_argument("--ddim_steps", type=int, default=50, help="steps of ddim if positive, otherwise use DDPM",) + parser.add_argument("--ddim_eta", type=float, default=1.0, help="eta for ddim sampling (0.0 yields deterministic sampling)",) + parser.add_argument("--bs", type=int, default=1, help="batch size for inference") + parser.add_argument("--height", type=int, default=512, help="image height, in pixel space") + parser.add_argument("--width", type=int, default=512, help="image width, in pixel space") + parser.add_argument("--unconditional_guidance_scale", type=float, default=1.0, help="prompt classifier-free guidance") + parser.add_argument("--unconditional_guidance_scale_temporal", type=float, default=None, help="temporal consistency guidance") + parser.add_argument("--seed", type=int, default=20230211, help="seed for seed_everything") + parser.add_argument("--cond_T", default=800, type=int, help="Steps smaller than cond_T will not contain condition") + parser.add_argument("--save_imgs", action='store_true', help="save condition") + parser.add_argument("--cond_dir", type=str, default=None, help="condition dir") + + return parser + + +if __name__ == '__main__': + now = datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S") + print("@CoLVDM cond-Inference: %s"%now) + parser = get_parser() + args, unkown = parser.parse_known_args() + # args = parser.parse_args() + + seed_everything(args.seed) + rank, gpu_num = 0, 1 + run_inference(args, gpu_num, rank) \ No newline at end of file diff --git a/main/evaluation/motionctrl_prompts_camerapose_trajs.py b/main/evaluation/motionctrl_prompts_camerapose_trajs.py new file mode 100644 index 0000000000000000000000000000000000000000..e668ef48ecdec41ed45efe3344ba695e63198c91 --- /dev/null +++ b/main/evaluation/motionctrl_prompts_camerapose_trajs.py @@ -0,0 +1,114 @@ +##### CMCM ##### +complex_camera_poses = [ + "test_camera_d971457c81bca597", + "test_camera_d971457c81bca597", + "test_camera_d971457c81bca597", + 'test_camera_Round-ZoomIn', + 'test_camera_Round-ZoomIn', + 'test_camera_Round-ZoomIn' +] +complex_camera_pose_prompt = [ + "a temple on a mountain, bird's view", + "Effiel Tower in Paris, bird's view", + "a castle in a forest, bird's view", + "a temple on a mountain, bird's view", + "Effiel Tower in Paris, bird's view", + "a castle in a forest, bird's view", + ] + +basic_camera_poses = [ + 'test_camera_L', + 'test_camera_D', + 'test_camera_I', + 'test_camera_O', + 'test_camera_R', + 'test_camera_U', + 'test_camera_SPIN-CW-60', + 'test_camera_SPIN-ACW-60', +] + +basic_camera_pose_prompt = [ + 'coastline, rocks, storm weather, wind, waves, lightning', + 'coastline, rocks, storm weather, wind, waves, lightning', + 'coastline, rocks, storm weather, wind, waves, lightning', + 'coastline, rocks, storm weather, wind, waves, lightning', + 'coastline, rocks, storm weather, wind, waves, lightning', + 'coastline, rocks, storm weather, wind, waves, lightning', + 'coastline, rocks, storm weather, wind, waves, lightning', + 'coastline, rocks, storm weather, wind, waves, lightning', +] + +diff_speeds_camera_poses = [ + 'test_camera_I_0.2x', + 'test_camera_I_0.4x', + 'test_camera_I_1.0x', + 'test_camera_I_2.0x', + + 'test_camera_O_0.2x', + 'test_camera_O_0.4x', + 'test_camera_O_1.0x', + 'test_camera_O_2.0x', +] + +diff_speeds_camera_pose_prompt = [ + 'A sunrise landscape features mountains and lakes', + 'A sunrise landscape features mountains and lakes', + 'A sunrise landscape features mountains and lakes', + 'A sunrise landscape features mountains and lakes', + 'A sunrise landscape features mountains and lakes', + 'A sunrise landscape features mountains and lakes', + 'A sunrise landscape features mountains and lakes', + 'A sunrise landscape features mountains and lakes', +] + +cmcm_prompt_camerapose = { + 'prompts': complex_camera_pose_prompt + basic_camera_pose_prompt + diff_speeds_camera_pose_prompt, + 'camera_poses': complex_camera_poses + basic_camera_poses + diff_speeds_camera_poses +} + +assert len(cmcm_prompt_camerapose['prompts']) == len(cmcm_prompt_camerapose['camera_poses']), \ + "The number of prompts and camera poses should be the same." + +### OMCM ### + +trajs = [ + 'shake_1', 'shake_1', 'shake_1', + 'curve_2', 'curve_2', 'curve_2', +] + +traj_prompt = [ + 'a sunflower swaying in the wind', + 'a rose swaying in the wind', + 'a wind chime swaying in the wind', + 'a man surfing', + 'a man skateboarding', + 'a girl skiing' +] + +omom_prompt_traj = { + 'prompts': traj_prompt, + 'trajs': trajs +} + +assert len(omom_prompt_traj['prompts']) == len(omom_prompt_traj['trajs']), \ + "The number of prompts and trajs should be the same." + +both_camerapose = [ + 'test_camera_O' +] +both_traj = [ + 'shaking_10' +] +both_prompt = [ + 'a rose swaying in the wind' +] + +both_prompt_camerapose_traj = { + 'prompts': both_prompt, + 'camera_poses': both_camerapose, + 'trajs': both_traj + +} + +assert len(both_prompt_camerapose_traj['prompts']) == len(both_prompt_camerapose_traj['camera_poses']) == len(both_prompt_camerapose_traj['trajs']), \ + "The number of prompts, camera poses and trajs should be the same." \ No newline at end of file diff --git a/motionctrl/lvdm_modified_modules.py b/motionctrl/lvdm_modified_modules.py new file mode 100644 index 0000000000000000000000000000000000000000..33c3480cdd8ac4d0cf8afd072c6436ccde4e3d8b --- /dev/null +++ b/motionctrl/lvdm_modified_modules.py @@ -0,0 +1,153 @@ +import logging + +import torch +from einops import rearrange, repeat + +from lvdm.models.utils_diffusion import timestep_embedding + +try: + import xformers + import xformers.ops + XFORMERS_IS_AVAILBLE = True +except: + XFORMERS_IS_AVAILBLE = False + +mainlogger = logging.getLogger('mainlogger') + + + +def TemporalTransformer_forward(self, x, context=None, is_imgbatch=False): + b, c, t, h, w = x.shape + x_in = x + x = self.norm(x) + x = rearrange(x, 'b c t h w -> (b h w) c t').contiguous() + if not self.use_linear: + x = self.proj_in(x) + x = rearrange(x, 'bhw c t -> bhw t c').contiguous() + if self.use_linear: + x = self.proj_in(x) + + temp_mask = None + if self.causal_attention: + temp_mask = torch.tril(torch.ones([1, t, t])) + if is_imgbatch: + temp_mask = torch.eye(t).unsqueeze(0) + if temp_mask is not None: + mask = temp_mask.to(x.device) + mask = repeat(mask, 'l i j -> (l bhw) i j', bhw=b*h*w) + else: + mask = None + + if self.only_self_att: + ## note: if no context is given, cross-attention defaults to self-attention + for i, block in enumerate(self.transformer_blocks): + x = block(x, context=context, mask=mask) + x = rearrange(x, '(b hw) t c -> b hw t c', b=b).contiguous() + else: + x = rearrange(x, '(b hw) t c -> b hw t c', b=b).contiguous() + context = rearrange(context, '(b t) l con -> b t l con', t=t).contiguous() + for i, block in enumerate(self.transformer_blocks): + # calculate each batch one by one (since number in shape could not greater then 65,535 for some package) + for j in range(b): + unit_context = context[j][0:1] + context_j = repeat(unit_context, 't l con -> (t r) l con', r=(h * w)).contiguous() + ## note: causal mask will not applied in cross-attention case + x[j] = block(x[j], context=context_j) + + if self.use_linear: + x = self.proj_out(x) + x = rearrange(x, 'b (h w) t c -> b c t h w', h=h, w=w).contiguous() + if not self.use_linear: + x = rearrange(x, 'b hw t c -> (b hw) c t').contiguous() + x = self.proj_out(x) + x = rearrange(x, '(b h w) c t -> b c t h w', b=b, h=h, w=w).contiguous() + + if self.use_image_dataset: + x = 0.0 * x + x_in + else: + x = x + x_in + return x + +def selfattn_forward_unet(self, x, timesteps, context=None, y=None, features_adapter=None, is_imgbatch=False, T=None, **kwargs): + b,_,t,_,_ = x.shape + + t_emb = timestep_embedding(timesteps, self.model_channels, repeat_only=False) + emb = self.time_embed(t_emb) + if self.micro_condition and y is not None: + micro_emb = timestep_embedding(y, self.model_channels, repeat_only=False) + emb = emb + self.micro_embed(micro_emb) + + + + # pose_emb = pose_emb.reshape(-1, pose_emb.shape[-1]) + ## repeat t times for context [(b t) 77 768] & time embedding + if not is_imgbatch: + context = context.repeat_interleave(repeats=t, dim=0) + + if 'pose_emb' in kwargs: + pose_emb = kwargs.pop('pose_emb') + context = { 'context': context, 'pose_emb': pose_emb } + + emb = emb.repeat_interleave(repeats=t, dim=0) + + ## always in shape (b t) c h w, except for temporal layer + x = rearrange(x, 'b c t h w -> (b t) c h w') + if features_adapter is not None: + features_adapter = [rearrange(feature, 'b c t h w -> (b t) c h w') for feature in features_adapter] + + h = x.type(self.dtype) + adapter_idx = 0 + hs = [] + for id, module in enumerate(self.input_blocks): + h = module(h, emb, context=context, batch_size=b,is_imgbatch=is_imgbatch) + if id ==0 and self.addition_attention: + h = self.init_attn(h, emb, context=context, batch_size=b,is_imgbatch=is_imgbatch) + ## plug-in adapter features + if ((id+1)%3 == 0) and features_adapter is not None: + # if adapter_idx == 0 or adapter_idx == 1 or adapter_idx == 2: + h = h + features_adapter[adapter_idx] + adapter_idx += 1 + hs.append(h) + if features_adapter is not None: + assert len(features_adapter)==adapter_idx, 'Wrong features_adapter' + + h = self.middle_block(h, emb, context=context, batch_size=b, is_imgbatch=is_imgbatch) + for module in self.output_blocks: + h = torch.cat([h, hs.pop()], dim=1) + h = module(h, emb, context=context, batch_size=b, is_imgbatch=is_imgbatch) + h = h.type(x.dtype) + y = self.out(h) + + # reshape back to (b c t h w) + y = rearrange(y, '(b t) c h w -> b c t h w', b=b) + return y + +def spatial_forward_BasicTransformerBlock(self, x, context=None, mask=None): + if isinstance(context, dict): + context = context['context'] + x = self.attn1(self.norm1(x), context=context if self.disable_self_attn else None, mask=mask) + x + x = self.attn2(self.norm2(x), context=context, mask=mask) + x + x = self.ff(self.norm3(x)) + x + return x + +def temporal_selfattn_forward_BasicTransformerBlock(self, x, context=None, mask=None): + if isinstance(context, dict) and 'pose_emb' in context: + pose_emb = context['pose_emb'] # {channel_num: [B, video_length, pose_dim, pose_embedding_dim]} + context = None + else: + pose_emb = None + context = None + + x = self.attn1(self.norm1(x), context=context if self.disable_self_attn else None, mask=mask) + x + + # Add camera pose + if pose_emb is not None: + B, t, _, _ = pose_emb.shape # [B, video_length, pose_dim, pose_embedding_dim] + hw = x.shape[0] // B + pose_emb = pose_emb.reshape(B, t, -1) + pose_emb = pose_emb.repeat_interleave(repeats=hw, dim=0) + x = self.cc_projection(torch.cat([x, pose_emb], dim=-1)) + + x = self.attn2(self.norm2(x), context=context, mask=mask) + x + x = self.ff(self.norm3(x)) + x + return x diff --git a/motionctrl/motionctrl.py b/motionctrl/motionctrl.py new file mode 100644 index 0000000000000000000000000000000000000000..bb8c6808f9ed450878d36054ac7082995c02ce6c --- /dev/null +++ b/motionctrl/motionctrl.py @@ -0,0 +1,67 @@ +import torch.nn as nn +from einops import rearrange + +from lvdm.models.ddpm3d import LatentDiffusion +from motionctrl.lvdm_modified_modules import ( + TemporalTransformer_forward, selfattn_forward_unet, + spatial_forward_BasicTransformerBlock, + temporal_selfattn_forward_BasicTransformerBlock) +from utils.utils import instantiate_from_config + + +class MotionCtrl(LatentDiffusion): + def __init__(self, + omcm_config=None, + pose_dim=12, + context_dim=1024, + *args, + **kwargs): + super(MotionCtrl, self).__init__(*args, **kwargs) + + # object motion control module + if omcm_config is not None: + self.omcm = instantiate_from_config(omcm_config) + else: + self.omcm = None + + + # camera motion control module + + bound_method = selfattn_forward_unet.__get__( + self.model.diffusion_model, + self.model.diffusion_model.__class__) + setattr(self.model.diffusion_model, 'forward', bound_method) + + for _name, _module in self.model.diffusion_model.named_modules(): + if _module.__class__.__name__ == 'TemporalTransformer': + bound_method = TemporalTransformer_forward.__get__( + _module, _module.__class__) + setattr(_module, 'forward', bound_method) + + if _module.__class__.__name__ == 'BasicTransformerBlock': + # SpatialTransformer only + if _module.attn2.to_k.in_features != context_dim: # TemporalTransformer without crossattn + + bound_method = temporal_selfattn_forward_BasicTransformerBlock.__get__( + _module, _module.__class__) + setattr(_module, '_forward', bound_method) + + cc_projection = nn.Linear(_module.attn2.to_k.in_features + pose_dim, _module.attn2.to_k.in_features) + nn.init.eye_(list(cc_projection.parameters())[0][:_module.attn2.to_k.in_features, :_module.attn2.to_k.in_features]) + nn.init.zeros_(list(cc_projection.parameters())[1]) + cc_projection.requires_grad_(True) + + _module.add_module('cc_projection', cc_projection) + + else: + bound_method = spatial_forward_BasicTransformerBlock.__get__( + _module, _module.__class__) + setattr(_module, '_forward', bound_method) + + def get_traj_features(self, extra_cond): + b, c, t, h, w = extra_cond.shape + ## process in 2D manner + extra_cond = rearrange(extra_cond, 'b c t h w -> (b t) c h w') + traj_features = self.omcm(extra_cond) + traj_features = [rearrange(feature, '(b t) c h w -> b c t h w', b=b, t=t) for feature in traj_features] + return traj_features diff --git a/requirements.txt b/requirements.txt new file mode 100644 index 0000000000000000000000000000000000000000..a561d88fab379f3c1daa6a926003999e34a7b271 --- /dev/null +++ b/requirements.txt @@ -0,0 +1,25 @@ +decord==0.6.0 +einops==0.3.0 +numpy==1.24.2 +omegaconf==2.1.1 +pandas==2.0.0 +Pillow==9.5.0 +pytorch_lightning==1.8.3 +PyYAML==6.0 +setuptools==65.6.3 +torch==2.0.0 +torchvision +tqdm==4.65.0 +transformers==4.25.1 +moviepy +av +xformers +timm +scikit-learn +open_clip_torch==2.12.0 +kornia +gradio==3.37.0 #3.35.2 +plotly +imageio==2.14.1 +imageio-ffmpeg==0.4.7 +opencv-python==4.8.0.74 \ No newline at end of file diff --git a/utils/utils.py b/utils/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..ac6d754b77cda98536b256d11dd40722f7ff8d2d --- /dev/null +++ b/utils/utils.py @@ -0,0 +1,80 @@ +import importlib +import numpy as np +from inspect import isfunction +from PIL import Image, ImageDraw, ImageFont +import cv2 + +import torch +import torch.distributed as dist + + +def count_params(model, verbose=False): + total_params = sum(p.numel() for p in model.parameters()) + if verbose: + print(f"{model.__class__.__name__} has {total_params*1.e-6:.2f} M params.") + return total_params + + +def check_istarget(name, para_list): + """ + name: full name of source para + para_list: partial name of target para + """ + istarget=False + for para in para_list: + if para in name: + return True + return istarget + + +def instantiate_from_config(config): + if not "target" in config: + if config == '__is_first_stage__': + return None + elif config == "__is_unconditional__": + return None + raise KeyError("Expected key `target` to instantiate.") + return get_obj_from_str(config["target"])(**config.get("params", dict())) + + +def get_obj_from_str(string, reload=False): + module, cls = string.rsplit(".", 1) + if reload: + module_imp = importlib.import_module(module) + importlib.reload(module_imp) + return getattr(importlib.import_module(module, package=None), cls) + + +def load_npz_from_dir(data_dir): + data = [np.load(os.path.join(data_dir, data_name))['arr_0'] for data_name in os.listdir(data_dir)] + data = np.concatenate(data, axis=0) + return data + + +def load_npz_from_paths(data_paths): + data = [np.load(data_path)['arr_0'] for data_path in data_paths] + data = np.concatenate(data, axis=0) + return data + + +def resize_numpy_image(image, max_resolution=512 * 512, resize_short_edge=None): + h, w = image.shape[:2] + if resize_short_edge is not None: + k = resize_short_edge / min(h, w) + else: + k = max_resolution / (h * w) + k = k**0.5 + h = int(np.round(h * k / 64)) * 64 + w = int(np.round(w * k / 64)) * 64 + image = cv2.resize(image, (w, h), interpolation=cv2.INTER_LANCZOS4) + return image + + +def setup_dist(args): + if dist.is_initialized(): + return + torch.cuda.set_device(args.local_rank) + torch.distributed.init_process_group( + 'nccl', + init_method='env://' + ) \ No newline at end of file