-
Notifications
You must be signed in to change notification settings - Fork 0
/
resnet50.py
166 lines (121 loc) · 4.95 KB
/
resnet50.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
#!/usr/bin/env python
import torch
import torch.nn as nn
import torch.nn.functional as F
import pdb
class IdBlock(nn.Module):
def __init__(self, in_channels, kernel_size, filters):
super(IdBlock, self).__init__()
self.in_channels = in_channels
self.k = kernel_size
self.f1, self.f2, self.f3 = filters
self.conv1 = nn.Conv2d(self.in_channels, self.f1, kernel_size=(1,1), stride=(1,1))
self.bn1 = nn.BatchNorm2d(self.f1)
self.conv2 = nn.Conv2d(self.f1, self.f2, kernel_size=(kernel_size,kernel_size), stride=(1,1), padding=self.k//2)
self.bn2 = nn.BatchNorm2d(self.f2)
self.conv3 = nn.Conv2d(self.f2, self.f3, kernel_size=(1,1), stride=(1,1))
self.bn3 = nn.BatchNorm2d(self.f3)
def forward(self, x):
shortcut = x
# first component of main path
out = self.conv1(x)
out = self.bn1(out)
out = F.relu(out)
# second component of main path
out = self.conv2(out)
out = self.bn2(out)
out = F.relu(out)
# third component of main path
out = self.conv3(out)
out = self.bn3(out)
out = out + shortcut
out = F.relu(out)
return out
class ConvBlock(nn.Module):
def __init__(self, in_channels, kernel_size, filters, stride=2):
super(ConvBlock, self).__init__()
self.in_channels = in_channels
self.k = kernel_size
self.f1, self.f2, self.f3 = filters
self.s = stride
self.conv1 = nn.Conv2d(self.in_channels, self.f1, kernel_size=(1,1), stride=(stride, stride))
self.bn1 = nn.BatchNorm2d(self.f1)
self.conv2 = nn.Conv2d(self.f1, self.f2, kernel_size=(kernel_size, kernel_size), stride=(1,1), padding = kernel_size//2)
self.bn2 = nn.BatchNorm2d(self.f2)
self.conv3 = nn.Conv2d(self.f2, self.f3, kernel_size=(1,1), stride=(1,1))
self.bn3 = nn.BatchNorm2d(self.f3)
self.convx = nn.Conv2d(self.in_channels, self.f3, kernel_size=(1,1), stride=(stride, stride))
self.bnx = nn.BatchNorm2d(self.f3)
def forward(self, x):
shortcut = x
# main path
out = self.conv1(x)
out = self.bn1(out)
out = F.relu(out)
# second component of main path
out = self.conv2(out)
out = self.bn2(out)
out = F.relu(out)
# third component of main path
out = self.conv3(out)
out = self.bn3(out)
# shortcut path
shortcut = self.convx(shortcut)
shortcut = self.bnx(shortcut)
out = out + shortcut
out = F.relu(out)
return out
class ResNet50(nn.Module):
def __init__(self, input_shape):
super(ResNet50, self).__init__()
self.classes = 13
self.conv1 = nn.Conv2d(input_shape[0], 64, 7, stride=(2,2), padding=3)
self.bn1 = nn.BatchNorm2d(64)
self.conv2_a = ConvBlock(64, kernel_size=3, filters=(64,64,256), stride = 1)
self.conv2_b = IdBlock(256, kernel_size=3, filters=(64,64,256))
self.conv2_c = IdBlock(256, kernel_size=3, filters=(64,64,256))
self.conv3_a = ConvBlock(256, kernel_size=3, filters=(128,128,512), stride=2)
self.conv3_b = IdBlock(512, kernel_size=3, filters=(128,128,512))
self.conv3_c = IdBlock(512, kernel_size=3, filters=(128,128,512))
self.conv3_d = IdBlock(512, kernel_size=3, filters=(128,128,512))
self.conv4_a = ConvBlock(512, kernel_size=3, filters=(256,256,1024), stride=2)
self.conv4_b = IdBlock(1024, kernel_size=3, filters=(256,256,1024))
self.conv4_c = IdBlock(1024, kernel_size=3, filters=(256,256,1024))
self.conv4_d = IdBlock(1024, kernel_size=3, filters=(256,256,1024))
self.conv4_e = IdBlock(1024, kernel_size=3, filters=(256,256,1024))
self.conv4_f = IdBlock(1024, kernel_size=3, filters=(256,256,1024))
self.conv5_a = ConvBlock(1024, kernel_size=3, filters=(512,512,2048), stride=2)
self.conv5_b = IdBlock(2048, kernel_size=3, filters=(512,512,2048))
self.conv5_c = IdBlock(2048, kernel_size=3, filters=(512,512,2048))
self.avgpool = nn.AdaptiveAvgPool2d((1,1))
self.fc1 = nn.Linear(2048, self.classes)
def forward(self, x):
# Stage 1
x = self.conv1(x)
x = self.bn1(x)
x = F.relu(x)
x = F.max_pool2d(x, kernel_size=3, stride=2)
# Stage 2
x = self.conv2_a(x)
x = self.conv2_b(x)
x = self.conv2_c(x)
# Stage 3
x = self.conv3_a(x)
x = self.conv3_b(x)
x = self.conv3_c(x)
x = self.conv3_d(x)
# Stage 4
x = self.conv4_a(x)
x = self.conv4_b(x)
x = self.conv4_c(x)
x = self.conv4_d(x)
x = self.conv4_e(x)
x = self.conv4_f(x)
# Stage 5
x = self.conv5_a(x)
x = self.conv5_b(x)
x = self.conv5_c(x)
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.fc1(x)
return x