-
Notifications
You must be signed in to change notification settings - Fork 0
/
data.py
149 lines (130 loc) · 4.32 KB
/
data.py
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
import os
import random
from pathlib import Path
import imageio
import numpy as np
import skimage.transform as sktr
import torch as th
import torchvision.transforms.functional as ttf
class Set68():
def __init__(
self,
color: bool = False,
device: th.device = th.device('cuda'),
):
n_ch = 3 if color else 1
self.images = th.empty((68, n_ch, 321, 481), device=device)
path = Path(os.environ['DATASETS_ROOT']) / 'set68'
for i, p in enumerate(sorted(path.iterdir())):
image = imageio.imread(str(p)) / 255
image = image if color else image.mean(-1, keepdims=True)
image = image if image.shape[0] == 321 else np.rot90(
image, axes=(0, 1)
)
self.images[i] = th.from_numpy(image.copy()).permute(2, 0, 1)
def data(self):
ims = ttf.center_crop(self.images, (320, 320))
return ims
class CelebA():
def __init__(
self,
color: bool = False,
size: int = 64,
device: th.device = th.device('cuda'),
n_im: int = 30_000,
batch_size: int = 64,
data: th.Tensor | None = None,
):
n_ch = 3 if color else 1
self.images = th.empty((n_im, n_ch, size, size), device=device)
self.batch_size = batch_size
self.n_im = n_im
if data is None:
path = Path(
os.environ['DATASETS_ROOT']
) / 'CelebAMask-HQ' / 'CelebA-HQ-img'
for i, p in enumerate(path.iterdir()):
print(i)
image = sktr.resize(
imageio.imread(str(p)) / 255,
output_shape=(size, size),
anti_aliasing=True,
)
if not color:
image = image.mean(-1, keepdims=True)
self.images[i] = th.from_numpy(image).permute(2, 0, 1)
if i == n_im - 1:
break
else:
self.images = data.clone()
def data(self, ) -> th.Tensor:
return self.images
def __iter__(self, ) -> th.Tensor:
while True:
ims = random.sample(list(range(self.n_im)), k=self.batch_size)
yield self.images[th.tensor(ims).long()]
class BSDS():
def __init__(
self,
color: bool = False,
batch_size: int = 50,
patch_size: int = 90,
rotate: bool = True,
flip: bool = True,
):
self.images = []
self.ch = 3 if color else 1
self.patch_size = patch_size
self.batch_size = batch_size
base = Path(os.environ['DATASETS_ROOT']) / 'bsds500'
for set in ['train', 'test']:
path = base / set
for im in path.iterdir():
image = imageio.imread(str(im)) / 255
if not color:
image = image.mean(-1, keepdims=True)
self.images.append(image)
self.flips = [
lambda x: x, lambda x: x[::-1, :, :], lambda x: x[:, ::-1, :],
lambda x: x[::-1, ::-1, :]
]
self.rng = np.random.default_rng(42)
self._transforms = [self.crop]
if rotate:
self._transforms.append(self.rotate)
if flip:
self._transforms.append(self.flip)
def transform(
self,
im: np.ndarray,
):
for tr in self._transforms:
im = tr(im)
return im
def rotate(
self,
im: np.ndarray,
) -> np.ndarray:
return np.rot90(im, self.rng.choice(4))
def flip(
self,
im: np.ndarray,
) -> np.ndarray:
return self.rng.choice(self.flips)(im)
def crop(
self,
im: np.ndarray,
) -> np.ndarray:
y = np.random.randint(0, im.shape[0] - self.patch_size)
x = np.random.randint(0, im.shape[1] - self.patch_size)
return im[y:y + self.patch_size, x:x + self.patch_size]
def __iter__(self, ) -> th.Tensor:
while True:
ims = random.choices(self.images, k=self.batch_size)
arr = np.empty(
(self.batch_size, self.ch, self.patch_size, self.patch_size)
)
for i, im in enumerate(ims):
arr[i] = self.transform(im).transpose(2, 0, 1)
ims = th.from_numpy(arr).cuda().float()
yield ims