-
Notifications
You must be signed in to change notification settings - Fork 2
/
ed_se_resnext.py
161 lines (129 loc) · 6.51 KB
/
ed_se_resnext.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
from __future__ import division
import torch.nn as nn
import torch.nn.functional as F
from torch.nn import init
import math
class Bottleneck(nn.Module):
expansion = 2
def __init__(self, in_channels, out_channels, stride, C, d, layer_idx=0):
super(Bottleneck, self).__init__()
conv1_in = in_channels
conv1_out = out_channels
conv2_in = conv1_out
conv2_out = conv2_in
conv3_in = conv2_out
conv3_out = out_channels*2
encoder_in = out_channels*2
encoder_out = out_channels
decoder_in = out_channels
decoder_out = out_channels*2
self.conv_conv1 = nn.Conv2d(conv1_in, conv1_out, kernel_size=1, stride=1, padding=0, bias=False)
self.bn_bn1 = nn.BatchNorm2d(conv1_out)
self.conv_conv2 = nn.Conv2d(conv2_in, conv2_out, kernel_size=3, stride=stride, padding=1, bias=False, groups=C)
self.bn_bn2 = nn.BatchNorm2d(conv2_out)
self.conv_conv3 = nn.Conv2d(conv3_in, conv3_out, kernel_size=1, stride=1 ,padding=0, bias=False)
self.bn_bn3 = nn.BatchNorm2d(conv3_out)
if out_channels==128:
self.GAPool = nn.AvgPool2d(56, stride=1)
elif out_channels==256:
self.GAPool = nn.AvgPool2d(28, stride=1)
elif out_channels==512:
self.GAPool = nn.AvgPool2d(14, stride=1)
elif out_channels==1024:
self.GAPool = nn.AvgPool2d(7, stride=1)
else:
print('GAPool Error\n')
assert 1==0
self.fc_reduction = nn.Linear(in_features=conv3_out, out_features=conv3_out//16)
self.fc_extention = nn.Linear(in_features=conv3_out//16, out_features=conv3_out)
self.sigmoid = nn.Sigmoid()
self.shortcut = nn.Sequential()
if (in_channels != out_channels * 2) or stride != 1:
self.shortcut.add_module('shortcut_conv',
nn.Conv2d(in_channels, out_channels*2, kernel_size=1, stride=stride, padding=0,
bias=False))
self.shortcut.add_module('shortcut_bn', nn.BatchNorm2d(out_channels*2))
self.encoder = nn.Sequential()
self.encoder.add_module('encoder_conv',
nn.Conv2d(encoder_in, encoder_out, kernel_size=3, stride=2, padding=1,groups=C, bias=False))
self.encoder.add_module('encoder_bn',
nn.BatchNorm2d(encoder_out))
self.decoder = nn.Sequential()
if layer_idx != 4:
self.decoder.add_module('decoder_conv',
nn.ConvTranspose2d(decoder_in,decoder_out,kernel_size=3,stride=2,padding=1,output_padding=1,groups=C,bias=False))
self.decoder.add_module('decoder_bn',
nn.BatchNorm2d(decoder_out))
else:
self.decoder.add_module('decoder_conv',
nn.ConvTranspose2d(decoder_in,decoder_out,kernel_size=3,stride=2,padding=1,output_padding=0,groups=C,bias=False))
self.decoder.add_module('decoder_bn',
nn.BatchNorm2d(decoder_out))
def forward(self, x):
proj = self.shortcut.forward(x)
encode = F.relu(self.encoder.forward(proj), inplace=True)
decode = self.decoder.forward(encode)
res = self.conv_conv1.forward(x)
res = F.relu(self.bn_bn1.forward(res), inplace=True)
res = self.conv_conv2.forward(res)
res = F.relu(self.bn_bn2.forward(res), inplace=True)
res = self.conv_conv3.forward(res)
res = self.bn_bn3.forward(res)
se_out = self.GAPool(res)
se_out = se_out.view(se_out.size(0), -1)
se_out = F.relu(self.fc_reduction(se_out), inplace=True)
se_out = self.fc_extention(se_out)
se_out = self.sigmoid(se_out)
se_out = se_out.view(se_out.size(0), se_out.size(1), 1, 1)
res = se_out*res
shtcut = self.shortcut.forward(x)
return F.relu(res + shtcut + decode, inplace=True)
class ResNeXt_SEED(nn.Module):
def __init__(self, block, C, d, layers, num_classes = 1000):
self.inplanes = 64
super(ResNeXt_SEED, self).__init__()
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer('layer1', block, planes=C*d*1, blocks=layers[0], stride=1, C=C, d=d, layer_idx=1)
self.layer2 = self._make_layer('layer2', block, planes=C*d*2, blocks=layers[1], stride=2, C=C, d=d, layer_idx=2)
self.layer3 = self._make_layer('layer3', block, planes=C*d*4, blocks=layers[2], stride=2, C=C, d=d, layer_idx=3)
self.layer4 = self._make_layer('layer4', block, planes=C*d*8, blocks=layers[3], stride=2, C=C, d=d, layer_idx=4)
self.avgpool = nn.AvgPool2d(7,stride=1)
self.fc = nn.Linear( (C*d*8) * block.expansion, num_classes)
for key in self.state_dict():
if key.split('.')[-1] == 'weight':
if 'conv' in key:
init.kaiming_normal(self.state_dict()[key], mode='fan_out')
if 'bn' in key:
self.state_dict()[key][...] = 1
elif key.split('.')[-1] == 'bias':
self.state_dict()[key][...] = 0
def _make_layer(self, name, block, planes, blocks, stride=1, C=32, d=4, layer_idx=0):#out_channel = planes
layers = nn.Sequential()
for block_idx in range(blocks):
name_ = '%s_block_%d' % (name, block_idx)
if block_idx == 0:
layers.add_module(name_, block(self.inplanes, planes, stride, C, d, layer_idx=layer_idx))
self.inplanes = planes * block.expansion
else:
layers.add_module(name_, block(self.inplanes, planes, 1, C, d, layer_idx=layer_idx))
return layers
def forward(self, x):
x = self.conv1.forward(x)
x = F.relu(self.bn1.forward(x), inplace=True)
x = self.maxpool(x)
x = self.layer1.forward(x)
x = self.layer2.forward(x)
x = self.layer3.forward(x)
x = self.layer4.forward(x)
x = self.avgpool(x)
x = x.view(x.size(0), -1)
x = self.fc(x)
return x
def se_resnext50_ed():
return ResNeXt_SEED(block=Bottleneck,C=32,d=4,layers=[3,4,6,3])
def se_resnext101_ed():
return ResNeXt_SEED(block=Bottleneck,C=32,d=4,layers=[3,4,23,3])
def se_resnext152_ed():
return ResNeXt_SEED(block=Bottleneck,C=32,d=4,layers=[3,8,36,3])