-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathdmvn.py
312 lines (225 loc) · 10.3 KB
/
dmvn.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
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
@author: liuyaqi
"""
import torch
import torch.nn as nn
import torchvision
import torch.nn.functional as F
use_bn = True
###############################################################################
def get_mapping_indices(n_rows, n_cols ):
new_indices = []
for r_x in range( n_rows ):
for c_x in range( n_cols ):
for r_b in range( n_rows ):
r_a = ( r_b + r_x ) % n_rows
for c_b in range( n_cols ):
c_a = ( c_b + c_x ) % n_cols
idx_a = r_a * n_cols + c_a
idx_b = r_b * n_cols + c_b
idx = idx_a * ( n_rows * n_cols ) + idx_b
new_indices.append( idx )
return new_indices
class Correlation_Module(nn.Module):
def __init__(self):
super(Correlation_Module, self).__init__()
def forward(self, x1, x2, new_indices):
[bs1, c1, h1, w1] = x1.size()
pixel_num = h1 * w1
x1 = torch.div(x1.view(bs1, c1, pixel_num),c1)
x2 = x2.view(bs1, c1, pixel_num).permute(0,2,1)
x1_x2 = torch.bmm(x2,x1)
x1_x2 = torch.index_select(x1_x2.view(bs1, -1),1,new_indices)
x1_x2 = x1_x2.view(bs1,pixel_num,h1,w1)
return x1_x2
class Poolopt_on_Corrmat(nn.Module):
def __init__(self, select_indices):
super(Poolopt_on_Corrmat, self).__init__()
self.select_indices = select_indices
for m in self.modules():
if isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
def forward(self, corr):
max_corr = torch.max(corr,1,keepdim=True)
avg_corr = torch.mean(corr,1,keepdim=True)
value1 = torch.sum(corr,3)
value = torch.sum(value1,2)
sort_v,sort_v_idx = torch.sort(value,1,descending=True)
sort_v_idx = torch.index_select(sort_v_idx,1,self.select_indices)
[bs,idx_num] = sort_v_idx.size()
for bs_idx in range(bs):
sort_corr_tmp = torch.index_select(torch.unsqueeze(corr[bs_idx],0),1,sort_v_idx[bs_idx])
if bs_idx == 0:
sort_corr = sort_corr_tmp
else:
sort_corr = torch.cat([sort_corr, sort_corr_tmp],0)
corr = torch.cat((max_corr[0],avg_corr, sort_corr),1)
return corr
###############################################################################
class Inception_pool(nn.Module):
def __init__(self, in_channels=8, nb_inceptors=32):
super(Inception_pool, self).__init__()
self.branch_pool = BasicConv2d(in_channels, nb_inceptors, kernel_size=1)
self.branch1x1 = BasicConv2d(in_channels, nb_inceptors, kernel_size=1)
self.branch3x3dbl_1 = BasicConv2d(in_channels, nb_inceptors, kernel_size=1)
self.branch3x3dbl_2 = BasicConv2d(nb_inceptors, nb_inceptors, kernel_size=3, padding=1)
self.branch5x5_1 = BasicConv2d(in_channels, nb_inceptors, kernel_size=1)
self.branch5x5_2 = BasicConv2d(nb_inceptors, nb_inceptors, kernel_size=5, padding=2)
self.final_conv = BasicConv2d(nb_inceptors*4, nb_inceptors, kernel_size=3, padding=1)
for m in self.modules():
if isinstance(m, nn.Conv2d) or isinstance(m, nn.Linear):
import scipy.stats as stats
stddev = m.stddev if hasattr(m, 'stddev') else 0.1
X = stats.truncnorm(-2, 2, scale=stddev)
values = torch.Tensor(X.rvs(m.weight.numel()))
values = values.view(m.weight.size())
m.weight.data.copy_(values)
if m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
def forward(self, x):
branch_pool = F.max_pool2d(x, kernel_size=3, stride=1, padding=1)
branch_pool = self.branch_pool(branch_pool)
branch1x1 = self.branch1x1(x)
branch3x3dbl = self.branch3x3dbl_1(x)
branch3x3dbl = self.branch3x3dbl_2(branch3x3dbl)
branch5x5 = self.branch5x5_1(x)
branch5x5 = self.branch5x5_2(branch5x5)
outputs = [branch_pool, branch1x1, branch3x3dbl, branch5x5]
final_output = torch.cat(outputs, 1)
final_output = self.final_conv(final_output)
return final_output
class BasicConv2d(nn.Module):
def __init__(self, in_channels, out_channels, **kwargs):
super(BasicConv2d, self).__init__()
self.conv = nn.Conv2d(in_channels, out_channels, bias=True, **kwargs)
if use_bn:
self.bn = nn.BatchNorm2d(out_channels, eps=0.001)
def forward(self, x):
x = self.conv(x)
if use_bn:
x = self.bn(x)
return F.relu(x, inplace=True)
class Inception_mask_small(nn.Module):
def __init__(self, in_channels=8, nb_inceptors=4):
super(Inception_mask_small, self).__init__()
self.branch1x1 = BasicConv2d(in_channels, nb_inceptors, kernel_size=1)
self.branch3x3 = BasicConv2d(in_channels, nb_inceptors, kernel_size=3, padding=1)
self.branch5x5 = BasicConv2d(in_channels, nb_inceptors, kernel_size=5, padding=2)
def forward(self, x):
branch1x1 = self.branch1x1(x)
branch3x3 = self.branch3x3(x)
branch5x5 = self.branch5x5(x)
outputs = [branch1x1, branch3x3, branch5x5]
final_output = torch.cat(outputs, 1)
return final_output
class Inception_mask_large(nn.Module):
def __init__(self, in_channels=8, nb_inceptors=4):
super(Inception_mask_large, self).__init__()
self.branch5x5 = BasicConv2d(in_channels, nb_inceptors, kernel_size=5, padding=2)
self.branch7x7 = BasicConv2d(in_channels, nb_inceptors, kernel_size=7, padding=3)
self.branch11x11 = BasicConv2d(in_channels, nb_inceptors, kernel_size=11, padding=5)
def forward(self, x):
branch5x5 = self.branch5x5(x)
branch7x7 = self.branch7x7(x)
branch11x11 = self.branch11x11(x)
outputs = [branch5x5, branch7x7, branch11x11]
final_output = torch.cat(outputs, 1)
return final_output
class mask_decoder(nn.Module):
def __init__(self, input_dim=8):
super(mask_decoder, self).__init__()
self.incep1 = Inception_mask_small(input_dim, nb_inceptors=4)
self.incep2 = Inception_mask_small(12, nb_inceptors=3)
self.incep3 = Inception_mask_small(9, nb_inceptors=2)
self.incep4 = Inception_mask_small(6, nb_inceptors=1)
self.incep5 = Inception_mask_large(3, nb_inceptors=1)
self.conv = nn.Conv2d(3, 1, 1, stride=1, padding=0)
for m in self.modules():
if isinstance(m, nn.Conv2d) or isinstance(m, nn.Linear):
import scipy.stats as stats
stddev = m.stddev if hasattr(m, 'stddev') else 0.1
X = stats.truncnorm(-2, 2, scale=stddev)
values = torch.Tensor(X.rvs(m.weight.numel()))
values = values.view(m.weight.size())
m.weight.data.copy_(values)
if m.bias is not None:
m.bias.data.zero_()
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
def forward(self, x):
x = F.upsample(x,size=(32,32),mode='bilinear',align_corners=True)
x = self.incep1(x)
x = F.upsample(x,size=(64,64),mode='bilinear',align_corners=True)
x = self.incep2(x)
x = F.upsample(x,size=(128,128),mode='bilinear',align_corners=True)
x = self.incep3(x)
x = F.upsample(x,size=(256,256),mode='bilinear',align_corners=True)
x = self.incep4(x)
x = self.incep5(x)
x = self.conv(x)
return x
#################################################
# VGG features
################################################
def base_feature():
vgg = torchvision.models.vgg16()
model = nn.Sequential()
[model.add_module(name=str(i), module=layer) for i, layer in enumerate(list(vgg.features)[:-7])]
return model
class dmvn(nn.Module):
def __init__(self,NoLabels,gpu_idx,h,w):
super(dmvn, self).__init__()
self.base = base_feature()
'''
Inception function module
'''
self.pool_inceptor = Inception_pool()
self.classifier = mask_decoder(64)
'''
The module used to measure the correspondence of two tensors
'''
self.gpu = gpu_idx
new_indices = get_mapping_indices(h, w)
self.new_indices = torch.tensor(new_indices,dtype=torch.long).cuda(self.gpu)
self.Corr = Correlation_Module()
'''
The corr maps pooling modules
'''
sort_indices = [0,1,2,3,4,5]
sort_indices = torch.tensor(sort_indices,dtype=torch.long).cuda(self.gpu)
self.poolopt_on_corrmat = Poolopt_on_Corrmat(sort_indices)
def forward(self,x1,x2):
# base feature extraction
x1 = self.base(x1)
x2 = self.base(x2)
# feature correlation
corr12 = self.Corr(x1,x2,self.new_indices)
corr21 = self.Corr(x2,x1,self.new_indices)
corr11 = self.Corr(x1,x1,self.new_indices)
corr22 = self.Corr(x2,x2,self.new_indices)
corr12 = self.poolopt_on_corrmat(corr12)
corr21 = self.poolopt_on_corrmat(corr21)
corr11 = self.poolopt_on_corrmat(corr11)
corr22 = self.poolopt_on_corrmat(corr22)
corr12 = self.pool_inceptor(corr12)
corr21 = self.pool_inceptor(corr21)
corr11 = self.pool_inceptor(corr11)
corr22 = self.pool_inceptor(corr22)
corr1 = torch.cat((corr12,corr11),1)
corr2 = torch.cat((corr21,corr22),1)
# output masks
mask1 = self.classifier(corr1)
mask2 = self.classifier(corr2)
return mask1, mask2
def DMVN_VGG(NoLabels,gpu_idx, dim):
w = dim/16
h = dim/16
model = dmvn(NoLabels,gpu_idx, h, w)
return model