Update animatediff/utils/util.py
Browse files- animatediff/utils/util.py +66 -0
animatediff/utils/util.py
CHANGED
@@ -9,6 +9,10 @@ import torchvision
|
|
9 |
from tqdm import tqdm
|
10 |
from einops import rearrange
|
11 |
|
|
|
|
|
|
|
|
|
12 |
|
13 |
def save_videos_grid(videos: torch.Tensor, path: str, rescale=False, n_rows=6, fps=8):
|
14 |
videos = rearrange(videos, "b c t h w -> t b c h w")
|
@@ -82,3 +86,65 @@ def ddim_loop(pipeline, ddim_scheduler, latent, num_inv_steps, prompt):
|
|
82 |
def ddim_inversion(pipeline, ddim_scheduler, video_latent, num_inv_steps, prompt=""):
|
83 |
ddim_latents = ddim_loop(pipeline, ddim_scheduler, video_latent, num_inv_steps, prompt)
|
84 |
return ddim_latents
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
9 |
from tqdm import tqdm
|
10 |
from einops import rearrange
|
11 |
|
12 |
+
import PIL.Image
|
13 |
+
import PIL.ImageOps
|
14 |
+
from packaging import version
|
15 |
+
from PIL import Image
|
16 |
|
17 |
def save_videos_grid(videos: torch.Tensor, path: str, rescale=False, n_rows=6, fps=8):
|
18 |
videos = rearrange(videos, "b c t h w -> t b c h w")
|
|
|
86 |
def ddim_inversion(pipeline, ddim_scheduler, video_latent, num_inv_steps, prompt=""):
|
87 |
ddim_latents = ddim_loop(pipeline, ddim_scheduler, video_latent, num_inv_steps, prompt)
|
88 |
return ddim_latents
|
89 |
+
|
90 |
+
if version.parse(version.parse(PIL.__version__).base_version) >= version.parse("9.1.0"):
|
91 |
+
PIL_INTERPOLATION = {
|
92 |
+
"linear": PIL.Image.Resampling.BILINEAR,
|
93 |
+
"bilinear": PIL.Image.Resampling.BILINEAR,
|
94 |
+
"bicubic": PIL.Image.Resampling.BICUBIC,
|
95 |
+
"lanczos": PIL.Image.Resampling.LANCZOS,
|
96 |
+
"nearest": PIL.Image.Resampling.NEAREST,
|
97 |
+
}
|
98 |
+
else:
|
99 |
+
PIL_INTERPOLATION = {
|
100 |
+
"linear": PIL.Image.LINEAR,
|
101 |
+
"bilinear": PIL.Image.BILINEAR,
|
102 |
+
"bicubic": PIL.Image.BICUBIC,
|
103 |
+
"lanczos": PIL.Image.LANCZOS,
|
104 |
+
"nearest": PIL.Image.NEAREST,
|
105 |
+
}
|
106 |
+
|
107 |
+
|
108 |
+
def pt_to_pil(images):
|
109 |
+
"""
|
110 |
+
Convert a torch image to a PIL image.
|
111 |
+
"""
|
112 |
+
images = (images / 2 + 0.5).clamp(0, 1)
|
113 |
+
images = images.cpu().permute(0, 2, 3, 1).float().numpy()
|
114 |
+
images = numpy_to_pil(images)
|
115 |
+
return images
|
116 |
+
|
117 |
+
|
118 |
+
def numpy_to_pil(images):
|
119 |
+
"""
|
120 |
+
Convert a numpy image or a batch of images to a PIL image.
|
121 |
+
"""
|
122 |
+
if images.ndim == 3:
|
123 |
+
images = images[None, ...]
|
124 |
+
images = (images * 255).round().astype("uint8")
|
125 |
+
if images.shape[-1] == 1:
|
126 |
+
# special case for grayscale (single channel) images
|
127 |
+
pil_images = [Image.fromarray(image.squeeze(), mode="L") for image in images]
|
128 |
+
else:
|
129 |
+
pil_images = [Image.fromarray(image) for image in images]
|
130 |
+
|
131 |
+
|
132 |
+
def preprocess_image(image):
|
133 |
+
if isinstance(image, torch.Tensor):
|
134 |
+
return image
|
135 |
+
elif isinstance(image, PIL.Image.Image):
|
136 |
+
image = [image]
|
137 |
+
|
138 |
+
if isinstance(image[0], PIL.Image.Image):
|
139 |
+
w, h = image[0].size
|
140 |
+
w, h = map(lambda x: x - x % 8, (w, h)) # resize to integer multiple of 8
|
141 |
+
|
142 |
+
image = [np.array(i.resize((w, h), resample=PIL_INTERPOLATION["lanczos"]))[None, :] for i in image]
|
143 |
+
image = np.concatenate(image, axis=0)
|
144 |
+
image = np.array(image).astype(np.float32) / 255.0
|
145 |
+
image = image.transpose(0, 3, 1, 2)
|
146 |
+
image = 2.0 * image - 1.0
|
147 |
+
image = torch.from_numpy(image)
|
148 |
+
elif isinstance(image[0], torch.Tensor):
|
149 |
+
image = torch.cat(image, dim=0)
|
150 |
+
return image
|