Spaces:
Sleeping
Sleeping
File size: 5,237 Bytes
e3af00f |
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 |
import asyncio
import math
from collections import deque
from concurrent.futures import ProcessPoolExecutor, ThreadPoolExecutor
from glob import glob
from pathlib import Path
import av
import numpy as np
from PIL import Image
from torch.utils.data import Dataset, default_collate
def get_default_video_reader(
data_path,
):
with av.open(str(data_path)) as container:
for frame in container.decode(video=0):
yield frame.to_ndarray(
format="rgb" if data_path.suffix == ".mp4" else "rgba"
)
accepted_format = set([".webp", ".png", ".jpg"])
def read_image(path):
return np.array(Image.open(path).convert("RGBA"))
class ImageDataset(Dataset):
def __init__(self, path, num_skip_frames=0):
paths = sorted(
[it for it in glob(f"{path}/*") if Path(it).suffix in accepted_format]
)
self.paths = paths[num_skip_frames:] + paths[:num_skip_frames]
def __getitem__(self, idx):
return read_image(self.paths[idx])
def __len__(self):
return len(self.paths)
class ProcessPoolIterator:
def __init__(self, dataset, preload=8, num_workers=2):
self.pool = ProcessPoolExecutor(num_workers)
self.dataset = dataset
self.queue = deque()
self.preload = preload
def __iter__(self):
for i in range(min(self.preload, len(self.dataset))):
self.queue.append(self.pool.submit(self.dataset.__getitem__, i))
for i in range(self.preload, len(self.dataset)):
self.queue.append(self.pool.submit(self.dataset.__getitem__, i))
yield self.queue.popleft().result()
while len(self.queue):
yield self.queue.popleft().result()
def __len__(self):
return len(self.dataset)
class ProcessPoolBatchIterator:
def __init__(self, dataset, batch_size, num_workers=4, drop_last=False):
self.iterator = ProcessPoolIterator(
dataset=dataset, preload=batch_size, num_workers=num_workers
)
self.batch_size = batch_size
self.drop_last = drop_last
def __iter__(self):
iterator = iter(self.iterator)
while True:
ret = []
try:
for i in range(self.batch_size):
ret.append(next(iterator))
yield default_collate(ret)
except StopIteration as e:
if not self.drop_last and ret:
yield default_collate(ret)
break
def __len__(self):
return (
math.floor(len(self.iterator) / self.batch_size)
if self.drop_last
else math.ceil(len(self.iterator) / self.batch_size)
)
class AsyncProcessPoolIterator:
def __init__(self, dataset, preload=8, num_workers=4):
self.pool = ProcessPoolExecutor(num_workers)
self.dataset = dataset
self.queue = deque()
self.preload = preload
async def __aiter__(self):
loop = asyncio.get_running_loop()
for i in range(min(self.preload, len(self.dataset))):
self.queue.append(
loop.run_in_executor(self.pool, self.dataset.__getitem__, i)
)
for i in range(self.preload, len(self.dataset)):
self.queue.append(
loop.run_in_executor(self.pool, self.dataset.__getitem__, i)
)
yield await self.queue.popleft()
while len(self.queue):
yield await self.queue.popleft()
def __len__(self):
return len(self.dataset)
class AsyncProcessPoolBatchIterator:
def __init__(self, dataset, batch_size, num_workers=4, drop_last=False):
self.iterator = AsyncProcessPoolIterator(
dataset=dataset, preload=batch_size, num_workers=num_workers
)
self.batch_size = batch_size
self.drop_last = drop_last
async def __aiter__(self):
iterator = aiter(self.iterator)
while True:
ret = []
try:
for _ in range(self.batch_size):
ret.append(await anext(iterator))
yield default_collate(ret)
except StopAsyncIteration as e:
if not self.drop_last and ret:
yield default_collate(ret)
break
def __len__(self):
return (
math.floor(len(self.iterator) / self.batch_size)
if self.drop_last
else math.ceil(len(self.iterator) / self.batch_size)
)
def get_image_folder_process_reader(
data_path,
num_skip_frames=0,
num_workers=4,
preload=16,
):
dataset = ImageDataset(path=data_path, num_skip_frames=num_skip_frames)
dataloader = ProcessPoolIterator(
dataset=dataset,
num_workers=num_workers,
preload=preload,
)
return dataloader
def get_image_folder_async_process_reader(
data_path,
num_skip_frames=0,
num_workers=4,
preload=16,
):
dataset = ImageDataset(path=data_path, num_skip_frames=num_skip_frames)
dataloader = AsyncProcessPoolIterator(
dataset=dataset,
num_workers=num_workers,
preload=preload,
)
return dataloader
|