-
Notifications
You must be signed in to change notification settings - Fork 16
/
preprocess.py
220 lines (174 loc) · 6.33 KB
/
preprocess.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
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
import torch
import torchvision.transforms as transforms
import random
__imagenet_stats = {'mean': [0.485, 0.456, 0.406],
'std': [0.229, 0.224, 0.225]}
__imagenet_pca = {
'eigval': torch.Tensor([0.2175, 0.0188, 0.0045]),
'eigvec': torch.Tensor([
[-0.5675, 0.7192, 0.4009],
[-0.5808, -0.0045, -0.8140],
[-0.5836, -0.6948, 0.4203],
])
}
def scale_crop(input_size, scale_size=None, normalize=__imagenet_stats):
t_list = [
transforms.CenterCrop(input_size),
transforms.ToTensor(),
transforms.Normalize(**normalize),
]
if scale_size != input_size:
t_list = [transforms.Scale(scale_size)] + t_list
return transforms.Compose(t_list)
def scale_random_crop(input_size, scale_size=None, normalize=__imagenet_stats):
t_list = [
transforms.RandomCrop(input_size),
transforms.ToTensor(),
transforms.Normalize(**normalize),
]
if scale_size != input_size:
t_list = [transforms.Scale(scale_size)] + t_list
transforms.Compose(t_list)
def scale_crop_nonorm(input_size):
t_list = [
transforms.CenterCrop(input_size),
transforms.ToTensor(),
]
return transforms.Compose(t_list)
def scale_random_nonorm( input_size ):
t_list = [
transforms.RandomCrop(input_size),
transforms.ToTensor(),
]
return transforms.Compose(t_list)
def pad_random_crop(input_size, scale_size=None, normalize=__imagenet_stats):
padding = int((scale_size - input_size) / 2)
return transforms.Compose([
transforms.RandomCrop(input_size, padding=padding),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(**normalize),
])
def inception_preproccess(input_size, normalize=__imagenet_stats):
return transforms.Compose([
transforms.RandomSizedCrop(input_size),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize(**normalize)
])
def inception_color_preproccess(input_size, normalize=__imagenet_stats):
return transforms.Compose([
transforms.RandomSizedCrop(input_size),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
ColorJitter(
brightness=0.4,
contrast=0.4,
saturation=0.4,
),
Lighting(0.1, __imagenet_pca['eigval'], __imagenet_pca['eigvec']),
transforms.Normalize(**normalize)
])
def get_transform(name='imagenet', input_size=None,
scale_size=None, normalize=None, augment=True):
normalize = normalize or __imagenet_stats
if name == 'imagenet':
scale_size = scale_size or 256
input_size = input_size or 224
if augment:
return inception_preproccess(input_size, normalize=normalize)
else:
return scale_crop(input_size=input_size,
scale_size=scale_size, normalize=normalize)
elif 'cifar' in name:
input_size = input_size or 32
if augment:
scale_size = scale_size or 40
return pad_random_crop(input_size, scale_size=scale_size,
normalize=normalize)
else:
scale_size = scale_size or 32
return scale_crop(input_size=input_size,
scale_size=scale_size, normalize=normalize)
elif name == 'mnist':
normalize = {'mean': [0.5], 'std': [0.5]}
input_size = input_size or 28
if augment:
scale_size = scale_size or 32
return pad_random_crop(input_size, scale_size=scale_size,
normalize=normalize)
else:
scale_size = scale_size or 32
return scale_crop(input_size=input_size,
scale_size=scale_size, normalize=normalize)
elif 'inria' in name:
input_size = input_size or 64
if augment:
return scale_random_nonorm(input_size=input_size)
else:
return scale_crop_nonorm(input_size=input_size)
class Lighting(object):
"""Lighting noise(AlexNet - style PCA - based noise)"""
def __init__(self, alphastd, eigval, eigvec):
self.alphastd = alphastd
self.eigval = eigval
self.eigvec = eigvec
def __call__(self, img):
if self.alphastd == 0:
return img
alpha = img.new().resize_(3).normal_(0, self.alphastd)
rgb = self.eigvec.type_as(img).clone()\
.mul(alpha.view(1, 3).expand(3, 3))\
.mul(self.eigval.view(1, 3).expand(3, 3))\
.sum(1).squeeze()
return img.add(rgb.view(3, 1, 1).expand_as(img))
class Grayscale(object):
def __call__(self, img):
gs = img.clone()
gs[0].mul_(0.299).add_(0.587, gs[1]).add_(0.114, gs[2])
gs[1].copy_(gs[0])
gs[2].copy_(gs[0])
return gs
class Saturation(object):
def __init__(self, var):
self.var = var
def __call__(self, img):
gs = Grayscale()(img)
alpha = random.uniform(0, self.var)
return img.lerp(gs, alpha)
class Brightness(object):
def __init__(self, var):
self.var = var
def __call__(self, img):
gs = img.new().resize_as_(img).zero_()
alpha = random.uniform(0, self.var)
return img.lerp(gs, alpha)
class Contrast(object):
def __init__(self, var):
self.var = var
def __call__(self, img):
gs = Grayscale()(img)
gs.fill_(gs.mean())
alpha = random.uniform(0, self.var)
return img.lerp(gs, alpha)
class RandomOrder(object):
""" Composes several transforms together in random order.
"""
def __init__(self, transforms):
self.transforms = transforms
def __call__(self, img):
if self.transforms is None:
return img
order = torch.randperm(len(self.transforms))
for i in order:
img = self.transforms[i](img)
return img
class ColorJitter(RandomOrder):
def __init__(self, brightness=0.4, contrast=0.4, saturation=0.4):
self.transforms = []
if brightness != 0:
self.transforms.append(Brightness(brightness))
if contrast != 0:
self.transforms.append(Contrast(contrast))
if saturation != 0:
self.transforms.append(Saturation(saturation))