Upload 6 files
Browse files- colorize.py +61 -0
- gradio_colorization.py +118 -0
- utils/data.py +47 -0
- utils/ddim.py +317 -0
- utils/diffusion.py +259 -0
- utils/model.py +43 -0
colorize.py
ADDED
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import random
|
2 |
+
|
3 |
+
import cv2
|
4 |
+
import einops
|
5 |
+
import numpy as np
|
6 |
+
import torch
|
7 |
+
from pytorch_lightning import seed_everything
|
8 |
+
|
9 |
+
from utils.data import HWC3, apply_color, resize_image
|
10 |
+
from utils.ddim import DDIMSampler
|
11 |
+
from utils.model import create_model, load_state_dict
|
12 |
+
|
13 |
+
model = create_model('./models/cldm_v21.yaml').cpu()
|
14 |
+
model.load_state_dict(load_state_dict(
|
15 |
+
'lightning_logs/version_6/checkpoints/colorizenet-sd21.ckpt', location='cuda'))
|
16 |
+
model = model.cuda()
|
17 |
+
ddim_sampler = DDIMSampler(model)
|
18 |
+
|
19 |
+
|
20 |
+
input_image = cv2.imread("sample_data/sample1_bw.jpg")
|
21 |
+
input_image = HWC3(input_image)
|
22 |
+
img = resize_image(input_image, resolution=512)
|
23 |
+
H, W, C = img.shape
|
24 |
+
|
25 |
+
num_samples = 1
|
26 |
+
control = torch.from_numpy(img.copy()).float().cuda() / 255.0
|
27 |
+
control = torch.stack([control for _ in range(num_samples)], dim=0)
|
28 |
+
control = einops.rearrange(control, 'b h w c -> b c h w').clone()
|
29 |
+
|
30 |
+
|
31 |
+
# seed = random.randint(0, 65535)
|
32 |
+
seed = 1294574436
|
33 |
+
seed_everything(seed)
|
34 |
+
prompt = "Colorize this image"
|
35 |
+
n_prompt = ""
|
36 |
+
guess_mode = False
|
37 |
+
strength = 1.0
|
38 |
+
eta = 0.0
|
39 |
+
ddim_steps = 20
|
40 |
+
scale = 9.0
|
41 |
+
|
42 |
+
cond = {"c_concat": [control], "c_crossattn": [
|
43 |
+
model.get_learned_conditioning([prompt] * num_samples)]}
|
44 |
+
un_cond = {"c_concat": None if guess_mode else [control], "c_crossattn": [
|
45 |
+
model.get_learned_conditioning([n_prompt] * num_samples)]}
|
46 |
+
shape = (4, H // 8, W // 8)
|
47 |
+
|
48 |
+
model.control_scales = [strength * (0.825 ** float(12 - i)) for i in range(13)] if guess_mode else (
|
49 |
+
[strength] * 13)
|
50 |
+
samples, intermediates = ddim_sampler.sample(ddim_steps, num_samples,
|
51 |
+
shape, cond, verbose=False, eta=eta,
|
52 |
+
unconditional_guidance_scale=scale,
|
53 |
+
unconditional_conditioning=un_cond)
|
54 |
+
|
55 |
+
x_samples = model.decode_first_stage(samples)
|
56 |
+
x_samples = (einops.rearrange(x_samples, 'b c h w -> b h w c')
|
57 |
+
* 127.5 + 127.5).cpu().numpy().clip(0, 255).astype(np.uint8)
|
58 |
+
|
59 |
+
results = [x_samples[i] for i in range(num_samples)]
|
60 |
+
colored_results = [apply_color(img, result) for result in results]
|
61 |
+
[cv2.imwrite(f"colorized_{i}.jpg", cv2.cvtColor(result, cv2.COLOR_RGB2BGR)) for i, result in enumerate(colored_results)]
|
gradio_colorization.py
ADDED
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from share import *
|
2 |
+
import config
|
3 |
+
|
4 |
+
import cv2
|
5 |
+
import einops
|
6 |
+
import gradio as gr
|
7 |
+
import numpy as np
|
8 |
+
import torch
|
9 |
+
import random
|
10 |
+
|
11 |
+
from pytorch_lightning import seed_everything
|
12 |
+
from annotator.util import resize_image, HWC3
|
13 |
+
from cldm.model import create_model, load_state_dict
|
14 |
+
from cldm.ddim_hacked import DDIMSampler
|
15 |
+
|
16 |
+
|
17 |
+
model = create_model('./models/cldm_v21.yaml').cpu()
|
18 |
+
model.load_state_dict(load_state_dict(
|
19 |
+
'lightning_logs/version_6/checkpoints/colorizenet-sd21.ckpt', location='cuda'))
|
20 |
+
model = model.cuda()
|
21 |
+
ddim_sampler = DDIMSampler(model)
|
22 |
+
|
23 |
+
|
24 |
+
def apply_color(image, color_map):
|
25 |
+
image = cv2.cvtColor(image, cv2.COLOR_RGB2LAB)
|
26 |
+
color_map = cv2.cvtColor(color_map, cv2.COLOR_RGB2LAB)
|
27 |
+
|
28 |
+
l, _, _ = cv2.split(image)
|
29 |
+
_, a, b = cv2.split(color_map)
|
30 |
+
|
31 |
+
merged = cv2.merge([l, a, b])
|
32 |
+
|
33 |
+
result = cv2.cvtColor(merged, cv2.COLOR_LAB2RGB)
|
34 |
+
return result
|
35 |
+
|
36 |
+
|
37 |
+
def process(input_image, prompt, a_prompt, n_prompt, num_samples, image_resolution, ddim_steps, guess_mode, strength, scale, seed, eta):
|
38 |
+
with torch.no_grad():
|
39 |
+
input_image = HWC3(input_image)
|
40 |
+
img = resize_image(input_image, image_resolution)
|
41 |
+
H, W, C = img.shape
|
42 |
+
|
43 |
+
control = torch.from_numpy(img.copy()).float().cuda() / 255.0
|
44 |
+
control = torch.stack([control for _ in range(num_samples)], dim=0)
|
45 |
+
control = einops.rearrange(control, 'b h w c -> b c h w').clone()
|
46 |
+
|
47 |
+
if seed == -1:
|
48 |
+
seed = random.randint(0, 65535)
|
49 |
+
seed_everything(seed)
|
50 |
+
|
51 |
+
if config.save_memory:
|
52 |
+
model.low_vram_shift(is_diffusing=False)
|
53 |
+
|
54 |
+
cond = {"c_concat": [control], "c_crossattn": [
|
55 |
+
model.get_learned_conditioning([prompt + ', ' + a_prompt] * num_samples)]}
|
56 |
+
un_cond = {"c_concat": None if guess_mode else [control], "c_crossattn": [
|
57 |
+
model.get_learned_conditioning([n_prompt] * num_samples)]}
|
58 |
+
shape = (4, H // 8, W // 8)
|
59 |
+
|
60 |
+
if config.save_memory:
|
61 |
+
model.low_vram_shift(is_diffusing=True)
|
62 |
+
|
63 |
+
model.control_scales = [strength * (0.825 ** float(12 - i)) for i in range(13)] if guess_mode else (
|
64 |
+
[strength] * 13) # Magic number. IDK why. Perhaps because 0.825**12<0.01 but 0.826**12>0.01
|
65 |
+
samples, intermediates = ddim_sampler.sample(ddim_steps, num_samples,
|
66 |
+
shape, cond, verbose=False, eta=eta,
|
67 |
+
unconditional_guidance_scale=scale,
|
68 |
+
unconditional_conditioning=un_cond)
|
69 |
+
|
70 |
+
if config.save_memory:
|
71 |
+
model.low_vram_shift(is_diffusing=False)
|
72 |
+
|
73 |
+
x_samples = model.decode_first_stage(samples)
|
74 |
+
x_samples = (einops.rearrange(x_samples, 'b c h w -> b h w c')
|
75 |
+
* 127.5 + 127.5).cpu().numpy().clip(0, 255).astype(np.uint8)
|
76 |
+
|
77 |
+
results = [x_samples[i] for i in range(num_samples)]
|
78 |
+
colored_results = [apply_color(img, result) for result in results]
|
79 |
+
return [img] + results + colored_results
|
80 |
+
|
81 |
+
|
82 |
+
block = gr.Blocks().queue()
|
83 |
+
with block:
|
84 |
+
with gr.Row():
|
85 |
+
gr.Markdown("## Colorize images with Stable Diffusion")
|
86 |
+
with gr.Row():
|
87 |
+
with gr.Column():
|
88 |
+
input_image = gr.Image(source='upload', type="numpy")
|
89 |
+
prompt = gr.Textbox(label="Prompt")
|
90 |
+
run_button = gr.Button(label="Run")
|
91 |
+
with gr.Accordion("Advanced options", open=False):
|
92 |
+
num_samples = gr.Slider(
|
93 |
+
label="Images", minimum=1, maximum=12, value=1, step=1)
|
94 |
+
image_resolution = gr.Slider(
|
95 |
+
label="Image Resolution", minimum=256, maximum=768, value=512, step=64)
|
96 |
+
strength = gr.Slider(
|
97 |
+
label="Control Strength", minimum=0.0, maximum=2.0, value=1.0, step=0.01)
|
98 |
+
guess_mode = gr.Checkbox(label='Guess Mode', value=False)
|
99 |
+
ddim_steps = gr.Slider(
|
100 |
+
label="Steps", minimum=1, maximum=100, value=20, step=1)
|
101 |
+
scale = gr.Slider(label="Guidance Scale",
|
102 |
+
minimum=0.1, maximum=30.0, value=9.0, step=0.1)
|
103 |
+
seed = gr.Slider(label="Seed", minimum=-1,
|
104 |
+
maximum=2147483647, step=1, randomize=True)
|
105 |
+
eta = gr.Number(label="eta (DDIM)", value=0.0)
|
106 |
+
a_prompt = gr.Textbox(
|
107 |
+
label="Added Prompt", value='best quality, natural colors')
|
108 |
+
n_prompt = gr.Textbox(label="Negative Prompt",
|
109 |
+
value='longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality')
|
110 |
+
with gr.Column():
|
111 |
+
result_gallery = gr.Gallery(
|
112 |
+
label='Output', show_label=False, elem_id="gallery").style(grid=3, height='auto')
|
113 |
+
ips = [input_image, prompt, a_prompt, n_prompt, num_samples, image_resolution,
|
114 |
+
ddim_steps, guess_mode, strength, scale, seed, eta]
|
115 |
+
run_button.click(fn=process, inputs=ips, outputs=[result_gallery])
|
116 |
+
|
117 |
+
|
118 |
+
block.launch(server_name='0.0.0.0')
|
utils/data.py
ADDED
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import cv2
|
2 |
+
import numpy as np
|
3 |
+
|
4 |
+
|
5 |
+
# Data utils
|
6 |
+
def HWC3(x):
|
7 |
+
assert x.dtype == np.uint8
|
8 |
+
if x.ndim == 2:
|
9 |
+
x = x[:, :, None]
|
10 |
+
assert x.ndim == 3
|
11 |
+
H, W, C = x.shape
|
12 |
+
assert C == 1 or C == 3 or C == 4
|
13 |
+
if C == 3:
|
14 |
+
return x
|
15 |
+
if C == 1:
|
16 |
+
return np.concatenate([x, x, x], axis=2)
|
17 |
+
if C == 4:
|
18 |
+
color = x[:, :, 0:3].astype(np.float32)
|
19 |
+
alpha = x[:, :, 3:4].astype(np.float32) / 255.0
|
20 |
+
y = color * alpha + 255.0 * (1.0 - alpha)
|
21 |
+
y = y.clip(0, 255).astype(np.uint8)
|
22 |
+
return y
|
23 |
+
|
24 |
+
|
25 |
+
def resize_image(input_image, resolution):
|
26 |
+
H, W, C = input_image.shape
|
27 |
+
H = float(H)
|
28 |
+
W = float(W)
|
29 |
+
k = float(resolution) / min(H, W)
|
30 |
+
H *= k
|
31 |
+
W *= k
|
32 |
+
H = int(np.round(H / 64.0)) * 64
|
33 |
+
W = int(np.round(W / 64.0)) * 64
|
34 |
+
img = cv2.resize(input_image, (W, H), interpolation=cv2.INTER_LANCZOS4 if k > 1 else cv2.INTER_AREA)
|
35 |
+
return img
|
36 |
+
|
37 |
+
def apply_color(image, color_map):
|
38 |
+
image = cv2.cvtColor(image, cv2.COLOR_RGB2LAB)
|
39 |
+
color_map = cv2.cvtColor(color_map, cv2.COLOR_RGB2LAB)
|
40 |
+
|
41 |
+
l, _, _ = cv2.split(image)
|
42 |
+
_, a, b = cv2.split(color_map)
|
43 |
+
|
44 |
+
merged = cv2.merge([l, a, b])
|
45 |
+
|
46 |
+
result = cv2.cvtColor(merged, cv2.COLOR_LAB2RGB)
|
47 |
+
return result
|
utils/ddim.py
ADDED
@@ -0,0 +1,317 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""SAMPLING ONLY."""
|
2 |
+
|
3 |
+
import torch
|
4 |
+
import numpy as np
|
5 |
+
from tqdm import tqdm
|
6 |
+
|
7 |
+
from ldm.modules.diffusionmodules.util import make_ddim_sampling_parameters, make_ddim_timesteps, noise_like, extract_into_tensor
|
8 |
+
|
9 |
+
|
10 |
+
class DDIMSampler(object):
|
11 |
+
def __init__(self, model, schedule="linear", **kwargs):
|
12 |
+
super().__init__()
|
13 |
+
self.model = model
|
14 |
+
self.ddpm_num_timesteps = model.num_timesteps
|
15 |
+
self.schedule = schedule
|
16 |
+
|
17 |
+
def register_buffer(self, name, attr):
|
18 |
+
if type(attr) == torch.Tensor:
|
19 |
+
if attr.device != torch.device("cuda"):
|
20 |
+
attr = attr.to(torch.device("cuda"))
|
21 |
+
setattr(self, name, attr)
|
22 |
+
|
23 |
+
def make_schedule(self, ddim_num_steps, ddim_discretize="uniform", ddim_eta=0., verbose=True):
|
24 |
+
self.ddim_timesteps = make_ddim_timesteps(ddim_discr_method=ddim_discretize, num_ddim_timesteps=ddim_num_steps,
|
25 |
+
num_ddpm_timesteps=self.ddpm_num_timesteps,verbose=verbose)
|
26 |
+
alphas_cumprod = self.model.alphas_cumprod
|
27 |
+
assert alphas_cumprod.shape[0] == self.ddpm_num_timesteps, 'alphas have to be defined for each timestep'
|
28 |
+
to_torch = lambda x: x.clone().detach().to(torch.float32).to(self.model.device)
|
29 |
+
|
30 |
+
self.register_buffer('betas', to_torch(self.model.betas))
|
31 |
+
self.register_buffer('alphas_cumprod', to_torch(alphas_cumprod))
|
32 |
+
self.register_buffer('alphas_cumprod_prev', to_torch(self.model.alphas_cumprod_prev))
|
33 |
+
|
34 |
+
# calculations for diffusion q(x_t | x_{t-1}) and others
|
35 |
+
self.register_buffer('sqrt_alphas_cumprod', to_torch(np.sqrt(alphas_cumprod.cpu())))
|
36 |
+
self.register_buffer('sqrt_one_minus_alphas_cumprod', to_torch(np.sqrt(1. - alphas_cumprod.cpu())))
|
37 |
+
self.register_buffer('log_one_minus_alphas_cumprod', to_torch(np.log(1. - alphas_cumprod.cpu())))
|
38 |
+
self.register_buffer('sqrt_recip_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu())))
|
39 |
+
self.register_buffer('sqrt_recipm1_alphas_cumprod', to_torch(np.sqrt(1. / alphas_cumprod.cpu() - 1)))
|
40 |
+
|
41 |
+
# ddim sampling parameters
|
42 |
+
ddim_sigmas, ddim_alphas, ddim_alphas_prev = make_ddim_sampling_parameters(alphacums=alphas_cumprod.cpu(),
|
43 |
+
ddim_timesteps=self.ddim_timesteps,
|
44 |
+
eta=ddim_eta,verbose=verbose)
|
45 |
+
self.register_buffer('ddim_sigmas', ddim_sigmas)
|
46 |
+
self.register_buffer('ddim_alphas', ddim_alphas)
|
47 |
+
self.register_buffer('ddim_alphas_prev', ddim_alphas_prev)
|
48 |
+
self.register_buffer('ddim_sqrt_one_minus_alphas', np.sqrt(1. - ddim_alphas))
|
49 |
+
sigmas_for_original_sampling_steps = ddim_eta * torch.sqrt(
|
50 |
+
(1 - self.alphas_cumprod_prev) / (1 - self.alphas_cumprod) * (
|
51 |
+
1 - self.alphas_cumprod / self.alphas_cumprod_prev))
|
52 |
+
self.register_buffer('ddim_sigmas_for_original_num_steps', sigmas_for_original_sampling_steps)
|
53 |
+
|
54 |
+
@torch.no_grad()
|
55 |
+
def sample(self,
|
56 |
+
S,
|
57 |
+
batch_size,
|
58 |
+
shape,
|
59 |
+
conditioning=None,
|
60 |
+
callback=None,
|
61 |
+
normals_sequence=None,
|
62 |
+
img_callback=None,
|
63 |
+
quantize_x0=False,
|
64 |
+
eta=0.,
|
65 |
+
mask=None,
|
66 |
+
x0=None,
|
67 |
+
temperature=1.,
|
68 |
+
noise_dropout=0.,
|
69 |
+
score_corrector=None,
|
70 |
+
corrector_kwargs=None,
|
71 |
+
verbose=True,
|
72 |
+
x_T=None,
|
73 |
+
log_every_t=100,
|
74 |
+
unconditional_guidance_scale=1.,
|
75 |
+
unconditional_conditioning=None, # this has to come in the same format as the conditioning, # e.g. as encoded tokens, ...
|
76 |
+
dynamic_threshold=None,
|
77 |
+
ucg_schedule=None,
|
78 |
+
**kwargs
|
79 |
+
):
|
80 |
+
if conditioning is not None:
|
81 |
+
if isinstance(conditioning, dict):
|
82 |
+
ctmp = conditioning[list(conditioning.keys())[0]]
|
83 |
+
while isinstance(ctmp, list): ctmp = ctmp[0]
|
84 |
+
cbs = ctmp.shape[0]
|
85 |
+
if cbs != batch_size:
|
86 |
+
print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
|
87 |
+
|
88 |
+
elif isinstance(conditioning, list):
|
89 |
+
for ctmp in conditioning:
|
90 |
+
if ctmp.shape[0] != batch_size:
|
91 |
+
print(f"Warning: Got {cbs} conditionings but batch-size is {batch_size}")
|
92 |
+
|
93 |
+
else:
|
94 |
+
if conditioning.shape[0] != batch_size:
|
95 |
+
print(f"Warning: Got {conditioning.shape[0]} conditionings but batch-size is {batch_size}")
|
96 |
+
|
97 |
+
self.make_schedule(ddim_num_steps=S, ddim_eta=eta, verbose=verbose)
|
98 |
+
# sampling
|
99 |
+
C, H, W = shape
|
100 |
+
size = (batch_size, C, H, W)
|
101 |
+
print(f'Data shape for DDIM sampling is {size}, eta {eta}')
|
102 |
+
|
103 |
+
samples, intermediates = self.ddim_sampling(conditioning, size,
|
104 |
+
callback=callback,
|
105 |
+
img_callback=img_callback,
|
106 |
+
quantize_denoised=quantize_x0,
|
107 |
+
mask=mask, x0=x0,
|
108 |
+
ddim_use_original_steps=False,
|
109 |
+
noise_dropout=noise_dropout,
|
110 |
+
temperature=temperature,
|
111 |
+
score_corrector=score_corrector,
|
112 |
+
corrector_kwargs=corrector_kwargs,
|
113 |
+
x_T=x_T,
|
114 |
+
log_every_t=log_every_t,
|
115 |
+
unconditional_guidance_scale=unconditional_guidance_scale,
|
116 |
+
unconditional_conditioning=unconditional_conditioning,
|
117 |
+
dynamic_threshold=dynamic_threshold,
|
118 |
+
ucg_schedule=ucg_schedule
|
119 |
+
)
|
120 |
+
return samples, intermediates
|
121 |
+
|
122 |
+
@torch.no_grad()
|
123 |
+
def ddim_sampling(self, cond, shape,
|
124 |
+
x_T=None, ddim_use_original_steps=False,
|
125 |
+
callback=None, timesteps=None, quantize_denoised=False,
|
126 |
+
mask=None, x0=None, img_callback=None, log_every_t=100,
|
127 |
+
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
|
128 |
+
unconditional_guidance_scale=1., unconditional_conditioning=None, dynamic_threshold=None,
|
129 |
+
ucg_schedule=None):
|
130 |
+
device = self.model.betas.device
|
131 |
+
b = shape[0]
|
132 |
+
if x_T is None:
|
133 |
+
img = torch.randn(shape, device=device)
|
134 |
+
else:
|
135 |
+
img = x_T
|
136 |
+
|
137 |
+
if timesteps is None:
|
138 |
+
timesteps = self.ddpm_num_timesteps if ddim_use_original_steps else self.ddim_timesteps
|
139 |
+
elif timesteps is not None and not ddim_use_original_steps:
|
140 |
+
subset_end = int(min(timesteps / self.ddim_timesteps.shape[0], 1) * self.ddim_timesteps.shape[0]) - 1
|
141 |
+
timesteps = self.ddim_timesteps[:subset_end]
|
142 |
+
|
143 |
+
intermediates = {'x_inter': [img], 'pred_x0': [img]}
|
144 |
+
time_range = reversed(range(0,timesteps)) if ddim_use_original_steps else np.flip(timesteps)
|
145 |
+
total_steps = timesteps if ddim_use_original_steps else timesteps.shape[0]
|
146 |
+
print(f"Running DDIM Sampling with {total_steps} timesteps")
|
147 |
+
|
148 |
+
iterator = tqdm(time_range, desc='DDIM Sampler', total=total_steps)
|
149 |
+
|
150 |
+
for i, step in enumerate(iterator):
|
151 |
+
index = total_steps - i - 1
|
152 |
+
ts = torch.full((b,), step, device=device, dtype=torch.long)
|
153 |
+
|
154 |
+
if mask is not None:
|
155 |
+
assert x0 is not None
|
156 |
+
img_orig = self.model.q_sample(x0, ts) # TODO: deterministic forward pass?
|
157 |
+
img = img_orig * mask + (1. - mask) * img
|
158 |
+
|
159 |
+
if ucg_schedule is not None:
|
160 |
+
assert len(ucg_schedule) == len(time_range)
|
161 |
+
unconditional_guidance_scale = ucg_schedule[i]
|
162 |
+
|
163 |
+
outs = self.p_sample_ddim(img, cond, ts, index=index, use_original_steps=ddim_use_original_steps,
|
164 |
+
quantize_denoised=quantize_denoised, temperature=temperature,
|
165 |
+
noise_dropout=noise_dropout, score_corrector=score_corrector,
|
166 |
+
corrector_kwargs=corrector_kwargs,
|
167 |
+
unconditional_guidance_scale=unconditional_guidance_scale,
|
168 |
+
unconditional_conditioning=unconditional_conditioning,
|
169 |
+
dynamic_threshold=dynamic_threshold)
|
170 |
+
img, pred_x0 = outs
|
171 |
+
if callback: callback(i)
|
172 |
+
if img_callback: img_callback(pred_x0, i)
|
173 |
+
|
174 |
+
if index % log_every_t == 0 or index == total_steps - 1:
|
175 |
+
intermediates['x_inter'].append(img)
|
176 |
+
intermediates['pred_x0'].append(pred_x0)
|
177 |
+
|
178 |
+
return img, intermediates
|
179 |
+
|
180 |
+
@torch.no_grad()
|
181 |
+
def p_sample_ddim(self, x, c, t, index, repeat_noise=False, use_original_steps=False, quantize_denoised=False,
|
182 |
+
temperature=1., noise_dropout=0., score_corrector=None, corrector_kwargs=None,
|
183 |
+
unconditional_guidance_scale=1., unconditional_conditioning=None,
|
184 |
+
dynamic_threshold=None):
|
185 |
+
b, *_, device = *x.shape, x.device
|
186 |
+
|
187 |
+
if unconditional_conditioning is None or unconditional_guidance_scale == 1.:
|
188 |
+
model_output = self.model.apply_model(x, t, c)
|
189 |
+
else:
|
190 |
+
model_t = self.model.apply_model(x, t, c)
|
191 |
+
model_uncond = self.model.apply_model(x, t, unconditional_conditioning)
|
192 |
+
model_output = model_uncond + unconditional_guidance_scale * (model_t - model_uncond)
|
193 |
+
|
194 |
+
if self.model.parameterization == "v":
|
195 |
+
e_t = self.model.predict_eps_from_z_and_v(x, t, model_output)
|
196 |
+
else:
|
197 |
+
e_t = model_output
|
198 |
+
|
199 |
+
if score_corrector is not None:
|
200 |
+
assert self.model.parameterization == "eps", 'not implemented'
|
201 |
+
e_t = score_corrector.modify_score(self.model, e_t, x, t, c, **corrector_kwargs)
|
202 |
+
|
203 |
+
alphas = self.model.alphas_cumprod if use_original_steps else self.ddim_alphas
|
204 |
+
alphas_prev = self.model.alphas_cumprod_prev if use_original_steps else self.ddim_alphas_prev
|
205 |
+
sqrt_one_minus_alphas = self.model.sqrt_one_minus_alphas_cumprod if use_original_steps else self.ddim_sqrt_one_minus_alphas
|
206 |
+
sigmas = self.model.ddim_sigmas_for_original_num_steps if use_original_steps else self.ddim_sigmas
|
207 |
+
# select parameters corresponding to the currently considered timestep
|
208 |
+
a_t = torch.full((b, 1, 1, 1), alphas[index], device=device)
|
209 |
+
a_prev = torch.full((b, 1, 1, 1), alphas_prev[index], device=device)
|
210 |
+
sigma_t = torch.full((b, 1, 1, 1), sigmas[index], device=device)
|
211 |
+
sqrt_one_minus_at = torch.full((b, 1, 1, 1), sqrt_one_minus_alphas[index],device=device)
|
212 |
+
|
213 |
+
# current prediction for x_0
|
214 |
+
if self.model.parameterization != "v":
|
215 |
+
pred_x0 = (x - sqrt_one_minus_at * e_t) / a_t.sqrt()
|
216 |
+
else:
|
217 |
+
pred_x0 = self.model.predict_start_from_z_and_v(x, t, model_output)
|
218 |
+
|
219 |
+
if quantize_denoised:
|
220 |
+
pred_x0, _, *_ = self.model.first_stage_model.quantize(pred_x0)
|
221 |
+
|
222 |
+
if dynamic_threshold is not None:
|
223 |
+
raise NotImplementedError()
|
224 |
+
|
225 |
+
# direction pointing to x_t
|
226 |
+
dir_xt = (1. - a_prev - sigma_t**2).sqrt() * e_t
|
227 |
+
noise = sigma_t * noise_like(x.shape, device, repeat_noise) * temperature
|
228 |
+
if noise_dropout > 0.:
|
229 |
+
noise = torch.nn.functional.dropout(noise, p=noise_dropout)
|
230 |
+
x_prev = a_prev.sqrt() * pred_x0 + dir_xt + noise
|
231 |
+
return x_prev, pred_x0
|
232 |
+
|
233 |
+
@torch.no_grad()
|
234 |
+
def encode(self, x0, c, t_enc, use_original_steps=False, return_intermediates=None,
|
235 |
+
unconditional_guidance_scale=1.0, unconditional_conditioning=None, callback=None):
|
236 |
+
timesteps = np.arange(self.ddpm_num_timesteps) if use_original_steps else self.ddim_timesteps
|
237 |
+
num_reference_steps = timesteps.shape[0]
|
238 |
+
|
239 |
+
assert t_enc <= num_reference_steps
|
240 |
+
num_steps = t_enc
|
241 |
+
|
242 |
+
if use_original_steps:
|
243 |
+
alphas_next = self.alphas_cumprod[:num_steps]
|
244 |
+
alphas = self.alphas_cumprod_prev[:num_steps]
|
245 |
+
else:
|
246 |
+
alphas_next = self.ddim_alphas[:num_steps]
|
247 |
+
alphas = torch.tensor(self.ddim_alphas_prev[:num_steps])
|
248 |
+
|
249 |
+
x_next = x0
|
250 |
+
intermediates = []
|
251 |
+
inter_steps = []
|
252 |
+
for i in tqdm(range(num_steps), desc='Encoding Image'):
|
253 |
+
t = torch.full((x0.shape[0],), timesteps[i], device=self.model.device, dtype=torch.long)
|
254 |
+
if unconditional_guidance_scale == 1.:
|
255 |
+
noise_pred = self.model.apply_model(x_next, t, c)
|
256 |
+
else:
|
257 |
+
assert unconditional_conditioning is not None
|
258 |
+
e_t_uncond, noise_pred = torch.chunk(
|
259 |
+
self.model.apply_model(torch.cat((x_next, x_next)), torch.cat((t, t)),
|
260 |
+
torch.cat((unconditional_conditioning, c))), 2)
|
261 |
+
noise_pred = e_t_uncond + unconditional_guidance_scale * (noise_pred - e_t_uncond)
|
262 |
+
|
263 |
+
xt_weighted = (alphas_next[i] / alphas[i]).sqrt() * x_next
|
264 |
+
weighted_noise_pred = alphas_next[i].sqrt() * (
|
265 |
+
(1 / alphas_next[i] - 1).sqrt() - (1 / alphas[i] - 1).sqrt()) * noise_pred
|
266 |
+
x_next = xt_weighted + weighted_noise_pred
|
267 |
+
if return_intermediates and i % (
|
268 |
+
num_steps // return_intermediates) == 0 and i < num_steps - 1:
|
269 |
+
intermediates.append(x_next)
|
270 |
+
inter_steps.append(i)
|
271 |
+
elif return_intermediates and i >= num_steps - 2:
|
272 |
+
intermediates.append(x_next)
|
273 |
+
inter_steps.append(i)
|
274 |
+
if callback: callback(i)
|
275 |
+
|
276 |
+
out = {'x_encoded': x_next, 'intermediate_steps': inter_steps}
|
277 |
+
if return_intermediates:
|
278 |
+
out.update({'intermediates': intermediates})
|
279 |
+
return x_next, out
|
280 |
+
|
281 |
+
@torch.no_grad()
|
282 |
+
def stochastic_encode(self, x0, t, use_original_steps=False, noise=None):
|
283 |
+
# fast, but does not allow for exact reconstruction
|
284 |
+
# t serves as an index to gather the correct alphas
|
285 |
+
if use_original_steps:
|
286 |
+
sqrt_alphas_cumprod = self.sqrt_alphas_cumprod
|
287 |
+
sqrt_one_minus_alphas_cumprod = self.sqrt_one_minus_alphas_cumprod
|
288 |
+
else:
|
289 |
+
sqrt_alphas_cumprod = torch.sqrt(self.ddim_alphas)
|
290 |
+
sqrt_one_minus_alphas_cumprod = self.ddim_sqrt_one_minus_alphas
|
291 |
+
|
292 |
+
if noise is None:
|
293 |
+
noise = torch.randn_like(x0)
|
294 |
+
return (extract_into_tensor(sqrt_alphas_cumprod, t, x0.shape) * x0 +
|
295 |
+
extract_into_tensor(sqrt_one_minus_alphas_cumprod, t, x0.shape) * noise)
|
296 |
+
|
297 |
+
@torch.no_grad()
|
298 |
+
def decode(self, x_latent, cond, t_start, unconditional_guidance_scale=1.0, unconditional_conditioning=None,
|
299 |
+
use_original_steps=False, callback=None):
|
300 |
+
|
301 |
+
timesteps = np.arange(self.ddpm_num_timesteps) if use_original_steps else self.ddim_timesteps
|
302 |
+
timesteps = timesteps[:t_start]
|
303 |
+
|
304 |
+
time_range = np.flip(timesteps)
|
305 |
+
total_steps = timesteps.shape[0]
|
306 |
+
print(f"Running DDIM Sampling with {total_steps} timesteps")
|
307 |
+
|
308 |
+
iterator = tqdm(time_range, desc='Decoding image', total=total_steps)
|
309 |
+
x_dec = x_latent
|
310 |
+
for i, step in enumerate(iterator):
|
311 |
+
index = total_steps - i - 1
|
312 |
+
ts = torch.full((x_latent.shape[0],), step, device=x_latent.device, dtype=torch.long)
|
313 |
+
x_dec, _ = self.p_sample_ddim(x_dec, cond, ts, index=index, use_original_steps=use_original_steps,
|
314 |
+
unconditional_guidance_scale=unconditional_guidance_scale,
|
315 |
+
unconditional_conditioning=unconditional_conditioning)
|
316 |
+
if callback: callback(i)
|
317 |
+
return x_dec
|
utils/diffusion.py
ADDED
@@ -0,0 +1,259 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import math
|
2 |
+
import torch
|
3 |
+
import torch.nn as nn
|
4 |
+
import numpy as np
|
5 |
+
from einops import repeat
|
6 |
+
|
7 |
+
from .model import instantiate_from_config
|
8 |
+
|
9 |
+
|
10 |
+
def make_beta_schedule(schedule, n_timestep, linear_start=1e-4, linear_end=2e-2, cosine_s=8e-3):
|
11 |
+
if schedule == "linear":
|
12 |
+
betas = (
|
13 |
+
torch.linspace(linear_start ** 0.5, linear_end ** 0.5, n_timestep, dtype=torch.float64) ** 2
|
14 |
+
)
|
15 |
+
|
16 |
+
elif schedule == "cosine":
|
17 |
+
timesteps = (
|
18 |
+
torch.arange(n_timestep + 1, dtype=torch.float64) / n_timestep + cosine_s
|
19 |
+
)
|
20 |
+
alphas = timesteps / (1 + cosine_s) * np.pi / 2
|
21 |
+
alphas = torch.cos(alphas).pow(2)
|
22 |
+
alphas = alphas / alphas[0]
|
23 |
+
betas = 1 - alphas[1:] / alphas[:-1]
|
24 |
+
betas = np.clip(betas, a_min=0, a_max=0.999)
|
25 |
+
|
26 |
+
elif schedule == "sqrt_linear":
|
27 |
+
betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64)
|
28 |
+
elif schedule == "sqrt":
|
29 |
+
betas = torch.linspace(linear_start, linear_end, n_timestep, dtype=torch.float64) ** 0.5
|
30 |
+
else:
|
31 |
+
raise ValueError(f"schedule '{schedule}' unknown.")
|
32 |
+
return betas.numpy()
|
33 |
+
|
34 |
+
|
35 |
+
def make_ddim_timesteps(ddim_discr_method, num_ddim_timesteps, num_ddpm_timesteps, verbose=True):
|
36 |
+
if ddim_discr_method == 'uniform':
|
37 |
+
c = num_ddpm_timesteps // num_ddim_timesteps
|
38 |
+
ddim_timesteps = np.asarray(list(range(0, num_ddpm_timesteps, c)))
|
39 |
+
elif ddim_discr_method == 'quad':
|
40 |
+
ddim_timesteps = ((np.linspace(0, np.sqrt(num_ddpm_timesteps * .8), num_ddim_timesteps)) ** 2).astype(int)
|
41 |
+
else:
|
42 |
+
raise NotImplementedError(f'There is no ddim discretization method called "{ddim_discr_method}"')
|
43 |
+
|
44 |
+
# assert ddim_timesteps.shape[0] == num_ddim_timesteps
|
45 |
+
# add one to get the final alpha values right (the ones from first scale to data during sampling)
|
46 |
+
steps_out = ddim_timesteps + 1
|
47 |
+
if verbose:
|
48 |
+
print(f'Selected timesteps for ddim sampler: {steps_out}')
|
49 |
+
return steps_out
|
50 |
+
|
51 |
+
|
52 |
+
def make_ddim_sampling_parameters(alphacums, ddim_timesteps, eta, verbose=True):
|
53 |
+
# select alphas for computing the variance schedule
|
54 |
+
alphas = alphacums[ddim_timesteps]
|
55 |
+
alphas_prev = np.asarray([alphacums[0]] + alphacums[ddim_timesteps[:-1]].tolist())
|
56 |
+
|
57 |
+
# according the the formula provided in https://arxiv.org/abs/2010.02502
|
58 |
+
sigmas = eta * np.sqrt((1 - alphas_prev) / (1 - alphas) * (1 - alphas / alphas_prev))
|
59 |
+
if verbose:
|
60 |
+
print(f'Selected alphas for ddim sampler: a_t: {alphas}; a_(t-1): {alphas_prev}')
|
61 |
+
print(f'For the chosen value of eta, which is {eta}, '
|
62 |
+
f'this results in the following sigma_t schedule for ddim sampler {sigmas}')
|
63 |
+
return sigmas, alphas, alphas_prev
|
64 |
+
|
65 |
+
|
66 |
+
def betas_for_alpha_bar(num_diffusion_timesteps, alpha_bar, max_beta=0.999):
|
67 |
+
"""
|
68 |
+
Create a beta schedule that discretizes the given alpha_t_bar function,
|
69 |
+
which defines the cumulative product of (1-beta) over time from t = [0,1].
|
70 |
+
:param num_diffusion_timesteps: the number of betas to produce.
|
71 |
+
:param alpha_bar: a lambda that takes an argument t from 0 to 1 and
|
72 |
+
produces the cumulative product of (1-beta) up to that
|
73 |
+
part of the diffusion process.
|
74 |
+
:param max_beta: the maximum beta to use; use values lower than 1 to
|
75 |
+
prevent singularities.
|
76 |
+
"""
|
77 |
+
betas = []
|
78 |
+
for i in range(num_diffusion_timesteps):
|
79 |
+
t1 = i / num_diffusion_timesteps
|
80 |
+
t2 = (i + 1) / num_diffusion_timesteps
|
81 |
+
betas.append(min(1 - alpha_bar(t2) / alpha_bar(t1), max_beta))
|
82 |
+
return np.array(betas)
|
83 |
+
|
84 |
+
|
85 |
+
def extract_into_tensor(a, t, x_shape):
|
86 |
+
b, *_ = t.shape
|
87 |
+
out = a.gather(-1, t)
|
88 |
+
return out.reshape(b, *((1,) * (len(x_shape) - 1)))
|
89 |
+
|
90 |
+
|
91 |
+
def checkpoint(func, inputs, params, flag):
|
92 |
+
"""
|
93 |
+
Evaluate a function without caching intermediate activations, allowing for
|
94 |
+
reduced memory at the expense of extra compute in the backward pass.
|
95 |
+
:param func: the function to evaluate.
|
96 |
+
:param inputs: the argument sequence to pass to `func`.
|
97 |
+
:param params: a sequence of parameters `func` depends on but does not
|
98 |
+
explicitly take as arguments.
|
99 |
+
:param flag: if False, disable gradient checkpointing.
|
100 |
+
"""
|
101 |
+
if flag:
|
102 |
+
args = tuple(inputs) + tuple(params)
|
103 |
+
return CheckpointFunction.apply(func, len(inputs), *args)
|
104 |
+
else:
|
105 |
+
return func(*inputs)
|
106 |
+
|
107 |
+
|
108 |
+
class CheckpointFunction(torch.autograd.Function):
|
109 |
+
@staticmethod
|
110 |
+
def forward(ctx, run_function, length, *args):
|
111 |
+
ctx.run_function = run_function
|
112 |
+
ctx.input_tensors = list(args[:length])
|
113 |
+
ctx.input_params = list(args[length:])
|
114 |
+
ctx.gpu_autocast_kwargs = {"enabled": torch.is_autocast_enabled(),
|
115 |
+
"dtype": torch.get_autocast_gpu_dtype(),
|
116 |
+
"cache_enabled": torch.is_autocast_cache_enabled()}
|
117 |
+
with torch.no_grad():
|
118 |
+
output_tensors = ctx.run_function(*ctx.input_tensors)
|
119 |
+
return output_tensors
|
120 |
+
|
121 |
+
@staticmethod
|
122 |
+
def backward(ctx, *output_grads):
|
123 |
+
ctx.input_tensors = [x.detach().requires_grad_(True) for x in ctx.input_tensors]
|
124 |
+
with torch.enable_grad(), \
|
125 |
+
torch.cuda.amp.autocast(**ctx.gpu_autocast_kwargs):
|
126 |
+
# Fixes a bug where the first op in run_function modifies the
|
127 |
+
# Tensor storage in place, which is not allowed for detach()'d
|
128 |
+
# Tensors.
|
129 |
+
shallow_copies = [x.view_as(x) for x in ctx.input_tensors]
|
130 |
+
output_tensors = ctx.run_function(*shallow_copies)
|
131 |
+
input_grads = torch.autograd.grad(
|
132 |
+
output_tensors,
|
133 |
+
ctx.input_tensors + ctx.input_params,
|
134 |
+
output_grads,
|
135 |
+
allow_unused=True,
|
136 |
+
)
|
137 |
+
del ctx.input_tensors
|
138 |
+
del ctx.input_params
|
139 |
+
del output_tensors
|
140 |
+
return (None, None) + input_grads
|
141 |
+
|
142 |
+
|
143 |
+
def timestep_embedding(timesteps, dim, max_period=10000, repeat_only=False):
|
144 |
+
"""
|
145 |
+
Create sinusoidal timestep embeddings.
|
146 |
+
:param timesteps: a 1-D Tensor of N indices, one per batch element.
|
147 |
+
These may be fractional.
|
148 |
+
:param dim: the dimension of the output.
|
149 |
+
:param max_period: controls the minimum frequency of the embeddings.
|
150 |
+
:return: an [N x dim] Tensor of positional embeddings.
|
151 |
+
"""
|
152 |
+
if not repeat_only:
|
153 |
+
half = dim // 2
|
154 |
+
freqs = torch.exp(
|
155 |
+
-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half
|
156 |
+
).to(device=timesteps.device)
|
157 |
+
args = timesteps[:, None].float() * freqs[None]
|
158 |
+
embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
|
159 |
+
if dim % 2:
|
160 |
+
embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
|
161 |
+
else:
|
162 |
+
embedding = repeat(timesteps, 'b -> b d', d=dim)
|
163 |
+
return embedding
|
164 |
+
|
165 |
+
|
166 |
+
def zero_module(module):
|
167 |
+
"""
|
168 |
+
Zero out the parameters of a module and return it.
|
169 |
+
"""
|
170 |
+
for p in module.parameters():
|
171 |
+
p.detach().zero_()
|
172 |
+
return module
|
173 |
+
|
174 |
+
|
175 |
+
def scale_module(module, scale):
|
176 |
+
"""
|
177 |
+
Scale the parameters of a module and return it.
|
178 |
+
"""
|
179 |
+
for p in module.parameters():
|
180 |
+
p.detach().mul_(scale)
|
181 |
+
return module
|
182 |
+
|
183 |
+
|
184 |
+
def mean_flat(tensor):
|
185 |
+
"""
|
186 |
+
Take the mean over all non-batch dimensions.
|
187 |
+
"""
|
188 |
+
return tensor.mean(dim=list(range(1, len(tensor.shape))))
|
189 |
+
|
190 |
+
|
191 |
+
def normalization(channels):
|
192 |
+
"""
|
193 |
+
Make a standard normalization layer.
|
194 |
+
:param channels: number of input channels.
|
195 |
+
:return: an nn.Module for normalization.
|
196 |
+
"""
|
197 |
+
return GroupNorm32(32, channels)
|
198 |
+
|
199 |
+
|
200 |
+
# PyTorch 1.7 has SiLU, but we support PyTorch 1.5.
|
201 |
+
class SiLU(nn.Module):
|
202 |
+
def forward(self, x):
|
203 |
+
return x * torch.sigmoid(x)
|
204 |
+
|
205 |
+
|
206 |
+
class GroupNorm32(nn.GroupNorm):
|
207 |
+
def forward(self, x):
|
208 |
+
return super().forward(x.float()).type(x.dtype)
|
209 |
+
|
210 |
+
def conv_nd(dims, *args, **kwargs):
|
211 |
+
"""
|
212 |
+
Create a 1D, 2D, or 3D convolution module.
|
213 |
+
"""
|
214 |
+
if dims == 1:
|
215 |
+
return nn.Conv1d(*args, **kwargs)
|
216 |
+
elif dims == 2:
|
217 |
+
return nn.Conv2d(*args, **kwargs)
|
218 |
+
elif dims == 3:
|
219 |
+
return nn.Conv3d(*args, **kwargs)
|
220 |
+
raise ValueError(f"unsupported dimensions: {dims}")
|
221 |
+
|
222 |
+
|
223 |
+
def linear(*args, **kwargs):
|
224 |
+
"""
|
225 |
+
Create a linear module.
|
226 |
+
"""
|
227 |
+
return nn.Linear(*args, **kwargs)
|
228 |
+
|
229 |
+
|
230 |
+
def avg_pool_nd(dims, *args, **kwargs):
|
231 |
+
"""
|
232 |
+
Create a 1D, 2D, or 3D average pooling module.
|
233 |
+
"""
|
234 |
+
if dims == 1:
|
235 |
+
return nn.AvgPool1d(*args, **kwargs)
|
236 |
+
elif dims == 2:
|
237 |
+
return nn.AvgPool2d(*args, **kwargs)
|
238 |
+
elif dims == 3:
|
239 |
+
return nn.AvgPool3d(*args, **kwargs)
|
240 |
+
raise ValueError(f"unsupported dimensions: {dims}")
|
241 |
+
|
242 |
+
|
243 |
+
class HybridConditioner(nn.Module):
|
244 |
+
|
245 |
+
def __init__(self, c_concat_config, c_crossattn_config):
|
246 |
+
super().__init__()
|
247 |
+
self.concat_conditioner = instantiate_from_config(c_concat_config)
|
248 |
+
self.crossattn_conditioner = instantiate_from_config(c_crossattn_config)
|
249 |
+
|
250 |
+
def forward(self, c_concat, c_crossattn):
|
251 |
+
c_concat = self.concat_conditioner(c_concat)
|
252 |
+
c_crossattn = self.crossattn_conditioner(c_crossattn)
|
253 |
+
return {'c_concat': [c_concat], 'c_crossattn': [c_crossattn]}
|
254 |
+
|
255 |
+
|
256 |
+
def noise_like(shape, device, repeat=False):
|
257 |
+
repeat_noise = lambda: torch.randn((1, *shape[1:]), device=device).repeat(shape[0], *((1,) * (len(shape) - 1)))
|
258 |
+
noise = lambda: torch.randn(shape, device=device)
|
259 |
+
return repeat_noise() if repeat else noise()
|
utils/model.py
ADDED
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from omegaconf import OmegaConf
|
2 |
+
from ldm.util import instantiate_from_config
|
3 |
+
import importlib
|
4 |
+
import os
|
5 |
+
import torch
|
6 |
+
|
7 |
+
|
8 |
+
def create_model(config_path):
|
9 |
+
config = OmegaConf.load(config_path)
|
10 |
+
model = instantiate_from_config(config.model).cpu()
|
11 |
+
print(f'Loaded model config from [{config_path}]')
|
12 |
+
return model
|
13 |
+
|
14 |
+
def instantiate_from_config(config):
|
15 |
+
if not "target" in config:
|
16 |
+
if config == '__is_first_stage__':
|
17 |
+
return None
|
18 |
+
elif config == "__is_unconditional__":
|
19 |
+
return None
|
20 |
+
raise KeyError("Expected key `target` to instantiate.")
|
21 |
+
return get_obj_from_str(config["target"])(**config.get("params", dict()))
|
22 |
+
|
23 |
+
|
24 |
+
def get_obj_from_str(string, reload=False):
|
25 |
+
module, cls = string.rsplit(".", 1)
|
26 |
+
if reload:
|
27 |
+
module_imp = importlib.import_module(module)
|
28 |
+
importlib.reload(module_imp)
|
29 |
+
return getattr(importlib.import_module(module, package=None), cls)
|
30 |
+
|
31 |
+
def get_state_dict(d):
|
32 |
+
return d.get('state_dict', d)
|
33 |
+
|
34 |
+
def load_state_dict(ckpt_path, location='cpu'):
|
35 |
+
_, extension = os.path.splitext(ckpt_path)
|
36 |
+
if extension.lower() == ".safetensors":
|
37 |
+
import safetensors.torch
|
38 |
+
state_dict = safetensors.torch.load_file(ckpt_path, device=location)
|
39 |
+
else:
|
40 |
+
state_dict = get_state_dict(torch.load(ckpt_path, map_location=torch.device(location)))
|
41 |
+
state_dict = get_state_dict(state_dict)
|
42 |
+
print(f'Loaded state_dict from [{ckpt_path}]')
|
43 |
+
return state_dict
|