Spaces:
Running
on
Zero
Running
on
Zero
File size: 4,244 Bytes
c42db24 |
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 |
import os
import torch
import torchvision
from PIL import Image
import numpy as np
import imageio
from einops import rearrange, repeat
def load_image(img, size):
# img = Image.open(filename).convert('RGB')
if not isinstance(img, np.ndarray):
img = Image.open(img).convert('RGB')
img = img.resize((size, size))
img = np.asarray(img)
img = np.transpose(img, (2, 0, 1)) # 3 x 256 x 256
return img / 255.0
def img_preprocessing(img_path, size):
img = load_image(img_path, size) # [0, 1]
img = torch.from_numpy(img).unsqueeze(0).float() # [0, 1]
imgs_norm = (img - 0.5) * 2.0 # [-1, 1]
return imgs_norm
def resize(img, size):
transform = torchvision.transforms.Compose([
torchvision.transforms.Resize(size, antialias=True),
torchvision.transforms.CenterCrop(size)
])
return transform(img)
def vid_preprocessing(vid_path, size):
vid_dict = torchvision.io.read_video(vid_path, pts_unit='sec')
vid = vid_dict[0].permute(0, 3, 1, 2).unsqueeze(0) # btchw
fps = vid_dict[2]['video_fps']
vid_norm = (vid / 255.0 - 0.5) * 2.0 # [-1, 1]
vid_norm = torch.cat([
resize(vid_norm[:, i, :, :, :], size).unsqueeze(1) for i in range(vid.size(1))
], dim=1)
return vid_norm, fps
def img_denorm(img):
img = img.clamp(-1, 1).cpu()
img = (img - img.min()) / (img.max() - img.min())
return img
def vid_denorm(vid):
vid = vid.clamp(-1, 1).cpu()
vid = (vid - vid.min()) / (vid.max() - vid.min())
return vid
def save_img_edit(save_dir, img, img_e):
# img: BCHW
# img_e: BCHW
output_img_path = os.path.join(save_dir, "img_edit.png")
output_img_all_path = os.path.join(save_dir, "img_all.png")
img = rearrange(img, 'b c h w -> b h w c')
img_e = rearrange(img_e, 'b c h w -> b h w c')
img_all = torch.cat([img, img_e], dim=2)
img_e_np = (img_denorm(img_e[0]).numpy() * 255).astype('uint8')
img_all_np = (img_denorm(img_all[0]).numpy() * 255).astype('uint8')
imageio.imwrite(output_img_path, img_e_np, quality=8)
imageio.imwrite(output_img_all_path, img_all_np, quality=8)
return
def save_vid_edit(save_dir, vid_d, vid_a, fps):
# img_s: BCHW
# vid_d: BTCHW
# vid_a: BCTHW
output_vid_a_path = os.path.join(save_dir, "vid_animation.mp4")
output_vid_all_path = os.path.join(save_dir, "vid_all.mp4")
vid_d = rearrange(vid_d, 'b t c h w -> b t h w c')
vid_a = rearrange(vid_a, 'b c t h w -> b t h w c')
vid_all = torch.cat([vid_d, vid_a], dim=3)
vid_a_np = (vid_denorm(vid_a[0]).numpy() * 255).astype('uint8')
vid_all_np = (vid_denorm(vid_all[0]).numpy() * 255).astype('uint8')
imageio.mimwrite(output_vid_a_path, vid_a_np, fps=fps, codec='libx264', quality=8)
imageio.mimwrite(output_vid_all_path, vid_all_np, fps=fps, codec='libx264', quality=8)
return
def save_animation(save_dir, img_s, vid_d, vid_a, fps):
# img_s: BCHW
# vid_d: BTCHW
# vid_a: BCTHW
output_vid_a_path = os.path.join(save_dir, "vid_animation.mp4")
output_img_e_path = os.path.join(save_dir, "img_edit.png")
output_vid_all_path = os.path.join(save_dir, "vid_all.mp4")
vid_d = rearrange(vid_d, 'b t c h w -> b t h w c')
vid_a = rearrange(vid_a, 'b c t h w -> b t h w c')
img_s = repeat(rearrange(img_s, 'b c h w -> b h w c'), 'b h w c -> b t h w c', t=vid_d.size(1))
vid_all = torch.cat([img_s, vid_d, vid_a], dim=3)
vid_a_np = (vid_denorm(vid_a[0]).numpy() * 255).astype('uint8')
img_e_np = vid_a_np[0]
vid_all_np = (vid_denorm(vid_all[0]).numpy() * 255).astype('uint8')
imageio.mimwrite(output_vid_a_path, vid_a_np, fps=fps, codec='libx264', quality=8)
imageio.mimwrite(output_vid_all_path, vid_all_np, fps=fps, codec='libx264', quality=8)
imageio.imwrite(output_img_e_path, img_e_np, quality=8)
return
def save_linear_manipulation(save_dir, vid, fps):
# vid: BCTHW
output_vid_path = os.path.join(save_dir, "vid_interpolation.mp4")
vid = rearrange(vid, 'b c t h w -> b t h w c')
vid_np = (vid_denorm(vid[0]).numpy() * 255).astype('uint8')
imageio.mimwrite(output_vid_path, vid_np, fps=fps, codec='libx264', quality=8)
return |