-
Notifications
You must be signed in to change notification settings - Fork 1
Expand file tree
/
Copy pathmodel.py
More file actions
162 lines (135 loc) · 6.42 KB
/
model.py
File metadata and controls
162 lines (135 loc) · 6.42 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
162
import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.nn.functional as F
from torch.nn import Parameter
import torchvision
import math
import scipy.io as sio
import numpy as np
class train_net(nn.Module):
def __init__(self):
super(train_net, self).__init__()
self.conv1_1 = nn.Conv2d(3, 64, 3, 1, 1)
self.relu1_1 = nn.ReLU()
self.conv1_2 = nn.Conv2d(64, 64, 3, 1, 1)
self.relu1_2 = nn.ReLU()
self.maxpool1 = nn.MaxPool2d(2, 2)
self.conv2_1 = nn.Conv2d(64, 128, 3, 1, 1)
self.relu2_1 = nn.ReLU()
self.conv2_2 = nn.Conv2d(128, 128, 3, 1, 1)
self.relu2_2 = nn.ReLU()
self.maxpool2 = nn.MaxPool2d(2, 2)
self.conv3_1 = nn.Conv2d(128, 256, 3, 1, 1)
self.relu3_1 = nn.ReLU()
self.conv3_2 = nn.Conv2d(256, 256, 3, 1, 1)
self.relu3_2 = nn.ReLU()
self.conv3_3 = nn.Conv2d(256, 256, 3, 1, 1)
self.relu3_3 = nn.ReLU()
self.maxpool3 = nn.MaxPool2d(2, 2)
self.conv4_1 = nn.Conv2d(256, 512, 3, 1, 1)
self.relu4_1 = nn.ReLU()
self.conv4_2 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu4_2 = nn.ReLU()
self.conv4_3 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu4_3 = nn.ReLU()
self.maxpool4 = nn.MaxPool2d(2, 2)
self.conv5_1 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu5_1 = nn.ReLU()
self.conv5_2 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu5_2 = nn.ReLU()
self.conv5_3 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu5_3 = nn.ReLU()
self.conv1_b1 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu1_b1 = nn.ReLU()
self.conv2_b1 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu2_b1 = nn.ReLU()
self.dropout_b1 = nn.Dropout(p=0.7) # Not sure about this layer
self.score_b1 = nn.Conv2d(512, 20, 1, 1, 0)
self.GAP_b1 = nn.AvgPool2d(14, 1) # Global Average Pooling
self.conv1_b2 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu1_b2 = nn.ReLU()
self.conv2_b2 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu2_b2 = nn.ReLU()
self.score_b2 = nn.Conv2d(512, 20, 1, 1, 0)
self.GAP_b2 = nn.AvgPool2d(14, 1) # Global Average Pooling
self.conv1_b3 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu1_b3 = nn.ReLU()
self.conv2_b3 = nn.Conv2d(512, 512, 3, 1, 1)
self.relu2_b3 = nn.ReLU()
self.score_b3 = nn.Conv2d(512, 20, 1, 1, 0)
self.GAP_b3 = nn.AvgPool2d(14, 1) # Global Average Pooling
def forward(self, x, labels):
x = self.maxpool1(self.relu1_2(self.conv1_2(self.relu1_1(self.conv1_1(x)))))
x = self.maxpool2(self.relu2_2(self.conv2_2(self.relu2_1(self.conv2_1(x)))))
x = self.maxpool3(self.relu3_3(self.conv3_3(self.relu3_2(self.conv3_2(self.relu3_1(self.conv3_1(x)))))))
x = self.maxpool4(self.relu4_3(self.conv4_3(self.relu4_2(self.conv4_2(self.relu4_1(self.conv4_1(x)))))))
features = self.relu5_3(self.conv5_3(self.relu5_2(self.conv5_2(self.relu5_1(self.conv5_1(x))))))
score_b1 = self.score_b1(self.dropout_b1(self.relu2_b1(self.conv2_b1(self.relu1_b1(self.conv1_b1(features))))))
prob1 = torch.sigmoid(self.GAP_b1(score_b1).squeeze())
features_b2 = mask_b2(features, score_b1, labels)
score_b2 = self.score_b2(self.relu2_b2(self.conv2_b2(self.relu1_b2(self.conv1_b2(features_b2)))))
prob2 = torch.sigmoid(self.GAP_b2(score_b2).squeeze())
features_b3 = mask_b3(features, score_b1, labels)
score_b3 = self.score_b3(self.relu2_b3(self.conv2_b3(self.relu1_b3(self.conv1_b3(features_b3)))))
prob3 = torch.sigmoid(self.GAP_b3(score_b3).squeeze())
return prob1, prob2, prob3, score_b1, score_b2
def mask_b2(features, score, labels, maxt=0.8, mint=0.05):
mask = score.clone()
mask[mask < 0] = 0
features = features.clone()
for i in range(20):
if torch.all(labels[:, i] == 0):
mask[:, i, :, :] = 0
else:
bs_label = torch.nonzero(labels[:, i]).reshape(-1)
ma, _ = mask[bs_label, i, :, :].reshape(-1, 14*14).max(dim=1)
ma = ma.reshape(-1, 1, 1)
mi, _ = mask[bs_label, i, :, :].reshape(-1, 14*14).min(dim=1)
mi = mi.reshape(-1, 1, 1)
tmp = (mask[bs_label, i, :, :] - mi) / (ma - mi + 1e-8)
mask[:, i, :, :] = 0
mask[bs_label, i, :, :] = tmp
mask, _ = mask.max(dim=1)
pos = torch.nonzero(mask > maxt).transpose(1, 0)
neg = torch.nonzero(mask < mint).transpose(1, 0)
features[pos[0], i, pos[1], pos[2]] = 0
features[neg[0], i, neg[1], neg[2]] = -1 * features[neg[0], i, neg[1], neg[2]]
return features
def mask_b3(features, score, labels, thres=0.3):
mask = score.clone()
mask[mask < 0] = 0
features = features.clone()
for i in range(20):
if torch.all(labels[:, i] == 0):
mask[:, i, :, :] = 0
else:
bs_label = torch.nonzero(labels[:, i]).reshape(-1)
ma, _ = mask[bs_label, i, :, :].reshape(-1, 14*14).max(dim=1)
ma = ma.reshape(-1, 1, 1)
mi, _ = mask[bs_label, i, :, :].reshape(-1, 14*14).min(dim=1)
mi = mi.reshape(-1, 1, 1)
tmp = (mask[bs_label, i, :, :] - mi) / (ma - mi + 1e-8)
mask[:, i, :, :] = 0
mask[bs_label, i, :, :] = tmp
mask, _ = mask.max(dim=1)
pos = torch.nonzero(mask > thres).transpose(1, 0)
features[pos[0], i, pos[1], pos[2]] = 0
return features
def load_vgg16pretrain(model, vggmodel='vgg16convs.mat'):
vgg16 = sio.loadmat(vggmodel)
torch_params = model.state_dict()
for k in vgg16.keys():
name_par = k.split('-')
size = len(name_par)
if size == 2:
name_space = name_par[0] + '.' + name_par[1]
data = np.squeeze(vgg16[k])
torch_params[name_space] = torch.from_numpy(data)
model.load_state_dict(torch_params)
def weights_init(m):
if isinstance(m, nn.Conv2d):
# xavier(m.weight.data)
m.weight.data.normal_(0, 0.01)
if m.bias is not None:
m.bias.data.zero_()