File size: 7,114 Bytes
b14067d ba201a1 b14067d ba201a1 b14067d ba201a1 b14067d ba201a1 b14067d ba201a1 b14067d fd926cd b14067d cdb41ad b14067d fd926cd b14067d ba201a1 b14067d ba201a1 b14067d fd926cd b14067d fd926cd b14067d fd926cd b14067d 0f464ea 43360f0 0f464ea 35ab635 b14067d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 |
from demo import GetAnchorVideos
import os
from datetime import datetime
import argparse
import torch
def get_parser():
parser = argparse.ArgumentParser()
## general
parser.add_argument('--video_path', type=str, help='Input path')
parser.add_argument(
'--out_dir', type=str, required=True, help='Output dir'
)
parser.add_argument(
'--device', type=str, default='cuda:0', help='The device to use'
)
parser.add_argument(
'--exp_name',
type=str,
default=None,
help='Experiment name, use video file name by default',
)
parser.add_argument(
'--save_name',
type=str,
default=None,
help='Experiment name, use video file name by default',
)
parser.add_argument(
'--seed', type=int, default=43, help='Random seed for reproducibility'
)
parser.add_argument(
'--video_length', type=int, default=49, help='Length of the video frames'
)
parser.add_argument('--fps', type=int, default=10, help='Fps for saved video')
parser.add_argument(
'--stride', type=int, default=1, help='Sampling stride for input video'
)
parser.add_argument('--server_name', type=str, help='Server IP address')
## render
parser.add_argument(
'--radius_scale',
type=float,
default=1.0,
help='Scale factor for the spherical radius',
)
parser.add_argument('--camera', type=str, default='traj', help='traj or target')
parser.add_argument(
'--mode', type=str, default='gradual', help='gradual, bullet or direct'
)
parser.add_argument(
'--mask', action='store_true', default=False, help='Clean the pcd if true'
)
parser.add_argument(
'--traj_txt',
type=str,
help="Required for 'traj' camera, a txt file that specify camera trajectory",
)
parser.add_argument(
'--target_pose',
nargs=5,
type=float,
help="Required for 'target' mode, specify target camera pose, <theta phi r x y>",
)
parser.add_argument(
'--near', type=float, default=0.0001, help='Near clipping plane distance'
)
parser.add_argument(
'--far', type=float, default=10000.0, help='Far clipping plane distance'
)
parser.add_argument(
'--height', type=int, default=1080, help='Height'
)
parser.add_argument(
'--width', type=int, default=1920, help='width'
)
parser.add_argument('--anchor_idx', type=int, default=0, help='One GT frame')
parser.add_argument(
'--near_far_estimated',
type=bool,
default=True,
help='Use estimated near and far values',
)
## diffusion
parser.add_argument(
'--low_gpu_memory_mode',
type=bool,
default=False,
help='Enable low GPU memory mode',
)
# parser.add_argument('--model_name', type=str, default='checkpoints/CogVideoX-Fun-V1.1-5b-InP', help='Path to the model')
parser.add_argument(
'--model_name',
type=str,
default='/app/pretrained/CogVideoX-Fun-V1.1-5b-InP',
help='Path to the model',
)
parser.add_argument(
'--sampler_name',
type=str,
choices=["Euler", "Euler A", "DPM++", "PNDM", "DDIM_Cog", "DDIM_Origin"],
default='DDIM_Origin',
help='Choose the sampler',
)
# parser.add_argument('--transformer_path', type=str,kdefault='checkpoints/TrajectoryCrafter/crosstransformer', help='Path to the pretrained transformer model')
parser.add_argument(
'--transformer_path',
type=str,
default="/app/pretrained/TrajectoryCrafter",
help='Path to the pretrained transformer model',
)
parser.add_argument(
'--sample_size',
type=int,
nargs=2,
default=[384, 672],
help='Sample size as [height, width]',
)
parser.add_argument(
'--diffusion_guidance_scale',
type=float,
default=6.0,
help='Guidance scale for inference',
)
parser.add_argument(
'--diffusion_inference_steps',
type=int,
default=50,
help='Number of inference steps',
)
parser.add_argument(
'--prompt', type=str, default=None, help='Prompt for video generation'
)
parser.add_argument(
'--negative_prompt',
type=str,
default="The video is not of a high quality, it has a low resolution. Watermark present in each frame. The background is solid.",
help='Negative prompt for video generation',
)
parser.add_argument(
'--refine_prompt',
type=str,
default=". The video is of high quality, and the view is very clear. ",
help='Prompt for video generation',
)
parser.add_argument('--qwen_path', type=str, default="/app/pretrained/Qwen2.5-VL-7B-Instruct")
## depth
# parser.add_argument('--unet_path', type=str, default='checkpoints/DepthCrafter', help='Path to the UNet model')
parser.add_argument(
'--unet_path',
type=str,
default="/app/pretrained/DepthCrafter",
help='Path to the UNet model',
)
# parser.add_argument('--pre_train_path', type=str, default='checkpoints/stable-video-diffusion-img2vid-xt', help='Path to the pre-trained model')
parser.add_argument(
'--pre_train_path',
type=str,
default="/app/pretrained/stable-video-diffusion-img2vid",
help='Path to the pre-trained model',
)
parser.add_argument(
'--cpu_offload', type=str, default='model', help='CPU offload strategy'
)
parser.add_argument(
'--depth_inference_steps', type=int, default=5, help='Number of inference steps'
)
parser.add_argument(
'--depth_guidance_scale',
type=float,
default=1.0,
help='Guidance scale for inference',
)
parser.add_argument(
'--window_size', type=int, default=110, help='Window size for processing'
)
parser.add_argument(
'--overlap', type=int, default=25, help='Overlap size for processing'
)
parser.add_argument(
'--max_res', type=int, default=1024, help='Maximum resolution for processing'
)
parser.add_argument("--target_aspect_ratio", type=int, nargs=2, default=None)
parser.add_argument('--init_dx', type=float, default=0.0)
parser.add_argument('--init_dy', type=float, default=0.0)
parser.add_argument('--init_dz', type=float, default=0.0)
return parser
if __name__ == "__main__":
parser = get_parser() # infer config.py
opts = parser.parse_args()
opts.weight_dtype = torch.bfloat16
pvd = GetAnchorVideos(opts)
if opts.mode == 'gradual':
pvd.infer_gradual(opts)
elif opts.mode == 'direct':
pvd.infer_direct(opts)
elif opts.mode == 'bullet':
pvd.infer_bullet(opts)
elif opts.mode == 'image':
pvd.infer_image(opts)
elif opts.mode == 'start_end':
pvd.infer_start_end(opts)
elif opts.mode == 'zoom':
pvd.infer_zoom(opts) |