transforms.py
3.79 KB
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
import numpy as np
import torch.nn as nn
import torchvision.transforms as transforms
from abc import ABC, abstractmethod
from PIL import Image, ImageOps, ImageEnhance
class BaseTransform(ABC):
def __init__(self, prob, mag):
self.prob = prob
self.mag = mag
def __call__(self, img):
return transforms.RandomApply([self.transform], self.prob)(img)
def __repr__(self):
return '%s(prob=%.2f, magnitude=%.2f)' % \
(self.__class__.__name__, self.prob, self.mag)
@abstractmethod
def transform(self, img):
pass
class ShearXY(BaseTransform):
def transform(self, img):
degrees = self.mag * 360
t = transforms.RandomAffine(0, shear=degrees, resample=Image.BILINEAR)
return t(img)
class TranslateXY(BaseTransform):
def transform(self, img):
translate = (self.mag, self.mag)
t = transforms.RandomAffine(0, translate=translate, resample=Image.BILINEAR)
return t(img)
class Rotate(BaseTransform):
def transform(self, img):
degrees = self.mag * 360
t = transforms.RandomRotation(degrees, Image.BILINEAR)
return t(img)
class AutoContrast(BaseTransform):
def transform(self, img):
cutoff = int(self.mag * 49)
return ImageOps.autocontrast(img, cutoff=cutoff)
class Invert(BaseTransform):
def transform(self, img):
return ImageOps.invert(img)
class Equalize(BaseTransform):
def transform(self, img):
return ImageOps.equalize(img)
class Solarize(BaseTransform):
def transform(self, img):
threshold = (1-self.mag) * 255
return ImageOps.solarize(img, threshold)
class Posterize(BaseTransform):
def transform(self, img):
bits = int((1-self.mag) * 8)
return ImageOps.posterize(img, bits=bits)
class Contrast(BaseTransform):
def transform(self, img):
factor = self.mag * 10
return ImageEnhance.Contrast(img).enhance(factor)
class Color(BaseTransform):
def transform(self, img):
factor = self.mag * 10
return ImageEnhance.Color(img).enhance(factor)
class Brightness(BaseTransform):
def transform(self, img):
factor = self.mag * 10
return ImageEnhance.Brightness(img).enhance(factor)
class Sharpness(BaseTransform):
def transform(self, img):
factor = self.mag * 10
return ImageEnhance.Sharpness(img).enhance(factor)
class Cutout(BaseTransform):
def transform(self, img):
n_holes = 1
length = 24 * self.mag
cutout_op = CutoutOp(n_holes=n_holes, length=length)
return cutout_op(img)
class CutoutOp(object):
"""
https://github.com/uoguelph-mlrg/Cutout
Randomly mask out one or more patches from an image.
Args:
n_holes (int): Number of patches to cut out of each image.
length (int): The length (in pixels) of each square patch.
"""
def __init__(self, n_holes, length):
self.n_holes = n_holes
self.length = length
def __call__(self, img):
"""
Args:
img (Tensor): Tensor image of size (C, H, W).
Returns:
Tensor: Image with n_holes of dimension length x length cut out of it.
"""
w, h = img.size
mask = np.ones((h, w, 1), np.uint8)
for n in range(self.n_holes):
y = np.random.randint(h)
x = np.random.randint(w)
y1 = np.clip(y - self.length // 2, 0, h).astype(int)
y2 = np.clip(y + self.length // 2, 0, h).astype(int)
x1 = np.clip(x - self.length // 2, 0, w).astype(int)
x2 = np.clip(x + self.length // 2, 0, w).astype(int)
mask[y1: y2, x1: x2, :] = 0.
img = mask*np.asarray(img).astype(np.uint8)
img = Image.fromarray(mask*np.asarray(img))
return img