-
Notifications
You must be signed in to change notification settings - Fork 6
/
vgg.py
143 lines (113 loc) · 4.49 KB
/
vgg.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
'''VGG11/13/16/19 in Pytorch.'''
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
cfg = {
'VGG11': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'VGG13': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
'VGG16': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'],
'VGG19': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'],
'VGG_ATT': [64, 64, 128, 128, 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M', 512, 'M', 512, 'M'],
}
class VGG(nn.Module):
def __init__(self, vgg_name):
super(VGG, self).__init__()
self.features = self._make_layers(cfg[vgg_name])
self.classifier = nn.Linear(512, 10)
def forward(self, x):
out = self.features(x)
out = out.view(out.size(0), -1)
out = self.classifier(out)
return out
def _make_layers(self, cfg):
layers = []
in_channels = 3
for x in cfg:
if x == 'M':
layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
else:
layers += [nn.Conv2d(in_channels, x, kernel_size=3, padding=1),
nn.BatchNorm2d(x),
nn.ReLU(inplace=True)]
in_channels = x
layers += [nn.AvgPool2d(kernel_size=1, stride=1)]
return nn.Sequential(*layers)
class VGG_ATT(nn.Module):
def __init__(self, mode='pc'):
super(VGG_ATT, self).__init__()
self.mode = mode
self.features = self._make_layers(cfg['VGG_ATT'])
self.classifier = nn.Linear(512, 10)
self.l1 = nn.Sequential(*list(self.features)[:22])
self.l2 = nn.Sequential(*list(self.features)[22:32])
self.l3 = nn.Sequential(*list(self.features)[32:42])
if mode == 'pc':
self.u1 = nn.Conv2d(256, 1, 1)
self.u2 = nn.Conv2d(512, 1, 1)
self.u3 = nn.Conv2d(512, 1, 1)
self.conv_out = nn.Sequential(*list(self.features)[42:50])
self.fc1 = nn.Linear(512, 512)
self.fc1_l1 = nn.Linear(512, 256)
self.fc1_l2 = nn.Linear(512, 512)
self.fc1_l3 = nn.Linear(512, 512)
self.fc2 = nn.Linear(256 + 512 + 512, 10)
def forward(self, x):
l1 = self.l1(x)
l2 = self.l2(l1)
l3 = self.l3(l2)
conv_out = self.conv_out(l3)
fc1 = self.fc1(conv_out.view(conv_out.size(0), -1))
fc1_l1 = self.fc1_l1(fc1)
fc1_l2 = self.fc1_l2(fc1)
fc1_l3 = self.fc1_l3(fc1)
att1 = self._compatibility_fn(l1, fc1_l1, level=1)
att2 = self._compatibility_fn(l2, fc1_l2, level=2)
att3 = self._compatibility_fn(l3, fc1_l3, level=3)
g1 = self._weighted_combine(l1, att1)
g2 = self._weighted_combine(l2, att2)
g3 = self._weighted_combine(l3, att3)
g = torch.cat((g1, g2, g3), dim=1)
out = self.fc2(g)
return out
def _make_layers(self, cfg):
layers = []
in_channels = 3
for x in cfg:
if x == 'M':
layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
else:
layers += [nn.Conv2d(in_channels, x, kernel_size=3, padding=1),
nn.BatchNorm2d(x),
nn.ReLU(inplace=True)]
in_channels = x
layers += [nn.AvgPool2d(kernel_size=1, stride=1)]
return nn.Sequential(*layers)
def _compatibility_fn(self, l, g, level):
if self.mode == 'dp':
att = l * g.unsqueeze(2).unsqueeze(3)
att = att.sum(1).unsqueeze(1)
size = att.size()
att = att.view(att.size(0), att.size(1), -1)
att = F.softmax(att, dim=2)
att = att.view(size)
elif self.mode == 'pc':
att = l + g.unsqueeze(2).unsqueeze(3)
if level == 1:
u = self.u1
elif level == 2:
u = self.u2
elif level == 3:
u = self.u3
att = u(att)
size = att.size()
att = att.view(att.size(0), att.size(1), -1)
att = F.softmax(att, dim=2)
att = att.view(size)
return att
def _weighted_combine(self, l, att_map):
g = l * att_map
return g.view(g.size(0), g.size(1), -1).sum(2)
# net = VGG_ATT()
# x = torch.randn(2,3,32,32)
# print(net(Variable(x)).size())