-
Notifications
You must be signed in to change notification settings - Fork 2
/
augment_edc.py
287 lines (251 loc) · 11.5 KB
/
augment_edc.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
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
# This code is based on the code provided in https://github.com/mit-han-lab/data-efficient-gans.
import torch
import torch.nn.functional as F
import numpy as np
class DiffAug():
def __init__(self,
strategy='color_crop_cutout_flip_scale_rotate',
batch=False,
ratio_cutout=0.5,
single=False):
self.prob_flip = 0.5
self.ratio_scale = 1.2
self.ratio_rotate = 15.0
self.ratio_crop_pad = 0.125
self.ratio_cutout = ratio_cutout
self.ratio_noise = 0.05
self.brightness = 1.0
self.saturation = 2.0
self.contrast = 0.5
self.batch = batch
self.aug = True
if strategy == '' or strategy.lower() == 'none':
self.aug = False
else:
self.strategy = []
self.flip = False
self.color = False
self.cutout = False
for aug in strategy.lower().split('_'):
if aug == 'flip' and single == False:
self.flip = True
elif aug == 'color' and single == False:
self.color = True
elif aug == 'cutout' and single == False:
self.cutout = True
else:
self.strategy.append(aug)
self.aug_fn = {
'color': [self.brightness_fn, self.saturation_fn, self.contrast_fn],
'crop': [self.crop_fn],
'cutout': [self.cutout_fn],
'flip': [self.flip_fn],
'scale': [self.scale_fn],
'rotate': [self.rotate_fn],
'translate': [self.translate_fn],
}
def __call__(self, x, single_aug=True, seed=-1):
if not self.aug:
return x
else:
if self.flip:
self.set_seed(seed)
x = self.flip_fn(x, self.batch)
if self.color:
for f in self.aug_fn['color']:
self.set_seed(seed)
x = f(x, self.batch)
if len(self.strategy) > 0:
if single_aug:
# single
idx = np.random.randint(len(self.strategy))
p = self.strategy[idx]
for f in self.aug_fn[p]:
self.set_seed(seed)
x = f(x, self.batch)
else:
# multiple
for p in self.strategy:
for f in self.aug_fn[p]:
self.set_seed(seed)
x = f(x, self.batch)
if self.cutout:
self.set_seed(seed)
x = self.cutout_fn(x, self.batch)
x = x.contiguous()
return x
def set_seed(self, seed):
if seed > 0:
np.random.seed(seed)
torch.random.manual_seed(seed)
def scale_fn(self, x, batch=True):
# x>1, max scale
# sx, sy: (0, +oo), 1: orignial size, 0.5: enlarge 2 times
ratio = self.ratio_scale
if batch:
sx = np.random.uniform() * (ratio - 1.0 / ratio) + 1.0 / ratio
sy = np.random.uniform() * (ratio - 1.0 / ratio) + 1.0 / ratio
theta = [[sx, 0, 0], [0, sy, 0]]
theta = torch.tensor(theta, dtype=torch.float, device=x.device)
theta = theta.expand(x.shape[0], 2, 3)
else:
sx = np.random.uniform(size=x.shape[0]) * (ratio - 1.0 / ratio) + 1.0 / ratio
sy = np.random.uniform(size=x.shape[0]) * (ratio - 1.0 / ratio) + 1.0 / ratio
theta = [[[sx[i], 0, 0], [0, sy[i], 0]] for i in range(x.shape[0])]
theta = torch.tensor(theta, dtype=torch.float, device=x.device)
grid = F.affine_grid(theta, x.shape)
x = F.grid_sample(x, grid)
return x
def rotate_fn(self, x, batch=True):
# [-180, 180], 90: anticlockwise 90 degree
ratio = self.ratio_rotate
if batch:
theta = (np.random.uniform() - 0.5) * 2 * ratio / 180 * float(np.pi)
theta = [[np.cos(theta), np.sin(-theta), 0], [np.sin(theta), np.cos(theta), 0]]
theta = torch.tensor(theta, dtype=torch.float, device=x.device)
theta = theta.expand(x.shape[0], 2, 3)
else:
theta = (np.random.uniform(size=x.shape[0]) - 0.5) * 2 * ratio / 180 * float(np.pi)
theta = [[[np.cos(theta[i]), np.sin(-theta[i]), 0],
[np.sin(theta[i]), np.cos(theta[i]), 0]] for i in range(x.shape[0])]
theta = torch.tensor(theta, dtype=torch.float, device=x.device)
grid = F.affine_grid(theta, x.shape)
x = F.grid_sample(x, grid)
return x
def flip_fn(self, x, batch=True):
prob = self.prob_flip
if batch:
coin = np.random.uniform()
if coin < prob:
return x.flip(3)
else:
return x
else:
randf = torch.rand(x.size(0), 1, 1, 1, device=x.device)
return torch.where(randf < prob, x.flip(3), x)
def brightness_fn(self, x, batch=True):
# mean
ratio = self.brightness
if batch:
randb = np.random.uniform()
else:
randb = torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device)
x = x + (randb - 0.5) * ratio
return x
def saturation_fn(self, x, batch=True):
# channel concentration
ratio = self.saturation
x_mean = x.mean(dim=1, keepdim=True)
if batch:
rands = np.random.uniform()
else:
rands = torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device)
x = (x - x_mean) * (rands * ratio) + x_mean
return x
def contrast_fn(self, x, batch=True):
# spatially concentrating
ratio = self.contrast
x_mean = x.mean(dim=[1, 2, 3], keepdim=True)
if batch:
randc = np.random.uniform()
else:
randc = torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device)
x = (x - x_mean) * (randc + ratio) + x_mean
return x
def translate_fn(self, x, batch=True):
ratio = self.ratio_crop_pad
shift_y = int(x.size(3) * ratio + 0.5)
if batch:
translation_y = np.random.randint(-shift_y, shift_y + 1)
else:
translation_y = torch.randint(-shift_y,
shift_y + 1,
size=[x.size(0), 1, 1],
device=x.device)
grid_batch, grid_x, grid_y = torch.meshgrid(
torch.arange(x.size(0), dtype=torch.long, device=x.device),
torch.arange(x.size(2), dtype=torch.long, device=x.device),
torch.arange(x.size(3), dtype=torch.long, device=x.device),
)
grid_y = torch.clamp(grid_y + translation_y + 1, 0, x.size(3) + 1)
x_pad = F.pad(x, (1, 1))
x = x_pad.permute(0, 2, 3, 1).contiguous()[grid_batch, grid_x, grid_y].permute(0, 3, 1, 2)
return x
def crop_fn(self, x, batch=True):
# The image is padded on its surrounding and then cropped.
ratio = self.ratio_crop_pad
shift_x, shift_y = int(x.size(2) * ratio + 0.5), int(x.size(3) * ratio + 0.5)
if batch:
translation_x = np.random.randint(-shift_x, shift_x + 1)
translation_y = np.random.randint(-shift_y, shift_y + 1)
else:
translation_x = torch.randint(-shift_x,
shift_x + 1,
size=[x.size(0), 1, 1],
device=x.device)
translation_y = torch.randint(-shift_y,
shift_y + 1,
size=[x.size(0), 1, 1],
device=x.device)
grid_batch, grid_x, grid_y = torch.meshgrid(
torch.arange(x.size(0), dtype=torch.long, device=x.device),
torch.arange(x.size(2), dtype=torch.long, device=x.device),
torch.arange(x.size(3), dtype=torch.long, device=x.device),
)
grid_x = torch.clamp(grid_x + translation_x + 1, 0, x.size(2) + 1)
grid_y = torch.clamp(grid_y + translation_y + 1, 0, x.size(3) + 1)
x_pad = F.pad(x, (1, 1, 1, 1))
x = x_pad.permute(0, 2, 3, 1).contiguous()[grid_batch, grid_x, grid_y].permute(0, 3, 1, 2)
return x
def cutout_fn(self, x, batch=True):
ratio = self.ratio_cutout
cutout_size = int(x.size(2) * ratio + 0.5), int(x.size(3) * ratio + 0.5)
if batch:
offset_x = np.random.randint(0, x.size(2) + (1 - cutout_size[0] % 2))
offset_y = np.random.randint(0, x.size(3) + (1 - cutout_size[1] % 2))
else:
offset_x = torch.randint(0,
x.size(2) + (1 - cutout_size[0] % 2),
size=[x.size(0), 1, 1],
device=x.device)
offset_y = torch.randint(0,
x.size(3) + (1 - cutout_size[1] % 2),
size=[x.size(0), 1, 1],
device=x.device)
grid_batch, grid_x, grid_y = torch.meshgrid(
torch.arange(x.size(0), dtype=torch.long, device=x.device),
torch.arange(cutout_size[0], dtype=torch.long, device=x.device),
torch.arange(cutout_size[1], dtype=torch.long, device=x.device),
)
grid_x = torch.clamp(grid_x + offset_x - cutout_size[0] // 2, min=0, max=x.size(2) - 1)
grid_y = torch.clamp(grid_y + offset_y - cutout_size[1] // 2, min=0, max=x.size(3) - 1)
mask = torch.ones(x.size(0), x.size(2), x.size(3), dtype=x.dtype, device=x.device)
mask[grid_batch, grid_x, grid_y] = 0
x = x * mask.unsqueeze(1)
return x
def cutout_inv_fn(self, x, batch=True):
ratio = self.ratio_cutout
cutout_size = int(x.size(2) * ratio + 0.5), int(x.size(3) * ratio + 0.5)
if batch:
offset_x = np.random.randint(0, x.size(2) - cutout_size[0])
offset_y = np.random.randint(0, x.size(3) - cutout_size[1])
else:
offset_x = torch.randint(0,
x.size(2) - cutout_size[0],
size=[x.size(0), 1, 1],
device=x.device)
offset_y = torch.randint(0,
x.size(3) - cutout_size[1],
size=[x.size(0), 1, 1],
device=x.device)
grid_batch, grid_x, grid_y = torch.meshgrid(
torch.arange(x.size(0), dtype=torch.long, device=x.device),
torch.arange(cutout_size[0], dtype=torch.long, device=x.device),
torch.arange(cutout_size[1], dtype=torch.long, device=x.device),
)
grid_x = torch.clamp(grid_x + offset_x, min=0, max=x.size(2) - 1)
grid_y = torch.clamp(grid_y + offset_y, min=0, max=x.size(3) - 1)
mask = torch.zeros(x.size(0), x.size(2), x.size(3), dtype=x.dtype, device=x.device)
mask[grid_batch, grid_x, grid_y] = 1.
x = x * mask.unsqueeze(1)
return x