Spaces:
Build error
Build error
File size: 11,368 Bytes
596242b |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 |
import os
import random
from typing import List
import av
import av.logging
import PIL
import numpy as np
from torch.utils.data import Dataset
from einops import rearrange
# from cache_decorator import Cache
import torch.nn.functional as F
import torchvision.transforms.functional as TF
import cv2
class VideoFolder(Dataset):
IMG_EXTENSIONS = [
".png",
".PNG",
".jpg",
".JPG"
]
VIDEO_EXTENSIONS = [".mp4", ".MP4", ".avi", ".AVI"]
def __init__(
self,
path: str,
size: List[int],
nframes: int = 128,
):
if isinstance(size, (list, tuple)):
if len(size) not in [1, 2]:
raise ValueError(
f"Size must be an int or a 1 or 2 element tuple/list, not a {len(size)} element tuple/list"
)
if isinstance(size, int):
size = [size, size]
def _find_all_path(_path):
_all_fnames = {
os.path.relpath(os.path.join(root, fname), start=_path)
for root, _dirs, files in os.walk(_path)
for fname in files
}
_video_fnames = sorted(
fname
for fname in _all_fnames
if self._file_ext(fname) in self.VIDEO_EXTENSIONS
) + sorted(
list(
set(
(
os.path.dirname(fname)
for fname in _all_fnames
if self._file_ext(fname) in self.IMG_EXTENSIONS
)
)
)
)
_video_fnames = sorted(_video_fnames)
return _video_fnames
_video_fnames = _find_all_path(path)
self.path = path
self.size = size
self.nframes = nframes
self._video_fnames = _video_fnames
self._total_size = len(self._video_fnames)
@staticmethod
def _file_ext(fname):
return os.path.splitext(fname)[1].lower()
def _read_video_opencv(self, video_path, nframes, size):
video = []
if os.path.isdir(video_path):
_all_fnames = {
os.path.relpath(os.path.join(root, fname), start=video_path)
for root, _dirs, files in os.walk(video_path)
for fname in files
}
_video_fnames = sorted(
fname
for fname in _all_fnames
if self._file_ext(fname) in self.IMG_EXTENSIONS
)
for fname in _video_fnames:
with open(os.path.join(video_path, fname), "rb") as f:
video.append(
np.array(
PIL.Image.open(f)
.convert("RGB")
.resize(
size, resample=3
) # PIL.Image.Resampling.LANCZOS = 1 PIL.Image.Resampling.BICUBIC = 3
)
)
else:
video = []
cap = cv2.VideoCapture(video_path)
while cap.isOpened():
success, image = cap.read()
if success:
video.append(
np.asarray(
cv2.resize(image, size, interpolation=cv2.INTER_CUBIC)[
:, :, ::-1
]
)
)
else:
break
cap.release()
if len(video) != nframes:
frame_scale = len(video) / nframes
frame_scaled_idxs = [int(i * frame_scale) for i in range(nframes)]
video = [video[i] for i in range(len(video)) if i in frame_scaled_idxs]
# for cache
video = np.stack(video).astype(np.uint8)
return video
# @Cache(
# cache_path="/data/kmei1/caches/{_hash}.pkl",
# )
def _read_video(self, video_path, nframes, size):
video = []
if os.path.isdir(video_path):
_all_fnames = {
os.path.relpath(os.path.join(root, fname), start=video_path)
for root, _dirs, files in os.walk(video_path)
for fname in files
}
_video_fnames = sorted(
[fname
for fname in _all_fnames
if self._file_ext(fname) in self.IMG_EXTENSIONS],
key = lambda x: int(x[:-4])
)
for fname in _video_fnames:
with open(os.path.join(video_path, fname), "rb") as f:
video.append(
np.array(
PIL.Image.open(f)
.convert("RGB")
.resize(
self.size, resample=1
) # PIL.Image.Resampling.LANCZOS = 1 PIL.Image.Resampling.BICUBIC = 3
)
)
else:
with av.open(video_path) as container:
container.streams.video[0].thread_type = "AUTO"
container.streams.video[0].thread_count = 2
total_frames = container.streams.video[0].frames
frame_scale = total_frames / nframes
frame_scaled_idxs = [int(i * frame_scale) for i in range(total_frames)]
for idx, frame in enumerate(container.decode(video=0)):
if idx in frame_scaled_idxs:
video.append(
np.asarray(
frame.to_image().resize(
size, resample=1
) # PIL.Image.Resampling.LANCZOS = 1 PIL.Image.Resampling.BICUBIC = 3
).clip(0, 255)
)
container.close()
frame_scale = len(video) / nframes
frame_scaled_idxs = [int(i * frame_scale) for i in range(nframes)]
video = [video[i] for i in range(len(video)) if i in frame_scaled_idxs]
video = np.stack(video).astype(np.uint8) # for cache
return video
def _read_video_metric(self, video_path, nframes, size):
video = []
if os.path.isdir(video_path):
_all_fnames = {
os.path.relpath(os.path.join(root, fname), start=video_path)
for root, _dirs, files in os.walk(video_path)
for fname in files
}
_video_fnames = sorted(
fname
for fname in _all_fnames
if self._file_ext(fname) in self.IMG_EXTENSIONS
)
for fname in _video_fnames:
with open(os.path.join(video_path, fname), "rb") as f:
video.append(
np.array(
PIL.Image.open(f)
.convert("RGB")
.resize(
self.size, resample=1
) # PIL.Image.Resampling.LANCZOS = 1 PIL.Image.Resampling.BICUBIC = 3
)
)
else:
with av.open(video_path) as container:
container.streams.video[0].thread_type = "AUTO"
container.streams.video[0].thread_count = 2
total_frames = container.streams.video[0].frames
frame_scale = total_frames / nframes
frame_scaled_idxs = [int(i * frame_scale) for i in range(total_frames)]
for idx, frame in enumerate(container.decode(video=0)):
if idx in frame_scaled_idxs:
frame = F.interpolate(TF.pil_to_tensor(frame.to_image()).unsqueeze(0), size=size[0], mode='bilinear', align_corners=False)[0].numpy().clip(0, 255)
video.append(frame)
container.close()
frame_scale = len(video) / nframes
frame_scaled_idxs = [int(i * frame_scale) for i in range(nframes)]
video = [video[i] for i in range(len(video)) if i in frame_scaled_idxs]
# for cache
video = np.stack(video).astype(np.uint8)
return video
def __getitem__(self, index):
video_path = os.path.join(self.path, self._video_fnames[index])
try:
video = self._read_video_metric(
video_path=video_path, nframes=self.nframes, size=self.size
)
except Exception as e:
print("=> error with loading video", video_path, e)
video = self.__getitem__(index + 1)
if video.shape[0] != self.nframes: print("=> unconsisitent video frames", video_path, video.shape[0], "v.s.", self.nframes)
video = video.astype(np.float32)
video = (video - 127.5) / 127.5
return video
def __len__(self):
return self._total_size
class Dataset(VideoFolder):
def __init__(
self,
data_root: str,
resolution: List[int],
video_length: int = 128,
latent_scale = 8,
actions = 7,
):
super().__init__(data_root, size=resolution, nframes=video_length)
self.data_root = data_root
self.actions = actions
videos = os.listdir(data_root)
videos.sort()
videos = videos[:100_000]
self._video_fnames = videos
self._total_size = len(self._video_fnames)
self.latent_scale = latent_scale
def __getitem__(self, index):
index = index % self._total_size
video_path = os.path.join(self.data_root, self._video_fnames[index])
video = self._read_video(
video_path=video_path, nframes=self.nframes, size=self.size
)
video = video.astype(np.float32)
video = (video - 127.5) / 127.5
if video.shape[0] != self.nframes:
raise ValueError(
f"{video_path} has less than {self.nframes} frames only have {video.shape[0]}"
)
actions = []
with open(
os.path.join(self.data_root, self._video_fnames[index], "actions.txt"), "r"
) as f:
for line in f: actions.append(int(line.strip()))
# SIMPLE_MOVEMENT = [
# ['NOOP'],
# ['right'],
# ['right', 'A'],
# ['right', 'B'],
# ['right', 'A', 'B'],
# ['A'],
# ['left'],
# ]
video = rearrange(video, 'T H W C -> C T H W')
grid_size = [self.nframes, video.shape[2] // self.latent_scale, video.shape[3] // self.latent_scale, self.actions]
grid_t = np.arange(grid_size[0], dtype=np.float32)
grid_h = np.arange(grid_size[1], dtype=np.float32)
grid_w = np.arange(grid_size[2], dtype=np.float32)
grid_action = np.arange(grid_size[3], dtype=np.float32)
grid = np.meshgrid(grid_t, grid_h, grid_w, grid_action, indexing='ij') # here w goes first
grid = np.stack(grid, axis=0)
grid = rearrange(grid[:, np.arange(grid_size[0]), :, :, actions], "T N H W -> N T H W")
return (video, actions, grid)
def __len__(self):
return self._total_size
|