-
Notifications
You must be signed in to change notification settings - Fork 5
/
Complex_Google_Inception.py
160 lines (126 loc) · 5.95 KB
/
Complex_Google_Inception.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
"""
Complex Valued Google Inception From Scratch
Programmed by Mehdi Hosseini Moghadam
The Base Code Of Google Inception Has Been Taken From:
https://github.com/aladdinpersson
* MIT Licence
* 2022-02-15 Last Update
"""
import torch
from torch import nn
from complex_neural_net import *
class Complex_GoogLeNet(nn.Module):
def __init__(self, aux_logits=True, num_classes=1000):
super(Complex_GoogLeNet, self).__init__()
assert aux_logits == True or aux_logits == False
self.aux_logits = aux_logits
# Write in_channels, etc, all explicit in self.conv1, rest will write to
# make everything as compact as possible, kernel_size=3 instead of (3,3)
self.complex_conv1 = complex_conv_block(
in_channels=3,
out_channels=64,
kernel_size=(7, 7),
stride=(2, 2),
padding=(3, 3),
)
self.complex_maxpool1 = CMaxPool2d(kernel_size=3, stride=2, padding=1)
self.complex_conv2 = complex_conv_block(64, 192, kernel_size=3, stride=1, padding=1)
self.complex_maxpool2 = CMaxPool2d(kernel_size=3, stride=2, padding=1)
# In this order: in_channels, out_1x1, red_3x3, out_3x3, red_5x5, out_5x5, out_1x1pool
self.complex_inception3a = complex_Inception_block(192, 64, 96, 128, 16, 32, 32)
self.complex_inception3b = complex_Inception_block(256, 128, 128, 192, 32, 96, 64)
self.complex_maxpool3 = CMaxPool2d(kernel_size=(3, 3), stride=2, padding=1)
self.complex_inception4a = complex_Inception_block(480, 192, 96, 208, 16, 48, 64)
self.complex_inception4b = complex_Inception_block(512, 160, 112, 224, 24, 64, 64)
self.complex_inception4c = complex_Inception_block(512, 128, 128, 256, 24, 64, 64)
self.complex_inception4d = complex_Inception_block(512, 112, 144, 288, 32, 64, 64)
self.complex_inception4e = complex_Inception_block(528, 256, 160, 320, 32, 128, 128)
self.complex_maxpool4 = CMaxPool2d(kernel_size=3, stride=2, padding=1)
self.complex_inception5a = complex_Inception_block(832, 256, 160, 320, 32, 128, 128)
self.complex_inception5b = complex_Inception_block(832, 384, 192, 384, 48, 128, 128)
self.complex_avgpool = CAvgPool2d(kernel_size=7, stride=1)
self.dropout = nn.Dropout(p=0.4)
self.complex_linear = CLinear(1024, num_classes)
if self.aux_logits:
self.complex_aux1 = complex_InceptionAux(512, num_classes)
self.complex_aux2 = complex_InceptionAux(528, num_classes)
else:
self.complex_aux1 = self.complex_aux2 = None
def forward(self, x):
x = self.complex_conv1(x)
x = self.complex_maxpool1(x)
x = self.complex_conv2(x)
# x = self.conv3(x)
x = self.complex_maxpool2(x)
x = self.complex_inception3a(x)
x = self.complex_inception3b(x)
x = self.complex_maxpool3(x)
x = self.complex_inception4a(x)
# Auxiliary Softmax classifier 1
if self.aux_logits and self.training:
complex_aux1 = self.complex_aux1(x)
x = self.complex_inception4b(x)
x = self.complex_inception4c(x)
x = self.complex_inception4d(x)
# Auxiliary Softmax classifier 2
if self.aux_logits and self.training:
complex_aux2 = self.complex_aux2(x)
x = self.complex_inception4e(x)
x = self.complex_maxpool4(x)
x = self.complex_inception5a(x)
x = self.complex_inception5b(x)
x = self.complex_avgpool(x)
x = x.reshape(x.shape[0], -1, 2)
x = self.dropout(x)
x = self.complex_linear(x)
if self.aux_logits and self.training:
return complex_aux1, complex_aux2, x
else:
return x
class complex_Inception_block(nn.Module):
def __init__(
self, in_channels, out_1x1, red_3x3, out_3x3, red_5x5, out_5x5, out_1x1pool
):
super(complex_Inception_block, self).__init__()
self.complex_branch1 = complex_conv_block(in_channels, out_1x1, kernel_size=(1, 1))
self.complex_branch2 = nn.Sequential(
complex_conv_block(in_channels, red_3x3, kernel_size=(1, 1)),
complex_conv_block(red_3x3, out_3x3, kernel_size=(3, 3), padding=(1, 1)),
)
self.complex_branch3 = nn.Sequential(
complex_conv_block(in_channels, red_5x5, kernel_size=(1, 1)),
complex_conv_block(red_5x5, out_5x5, kernel_size=(5, 5), padding=(2, 2)),
)
self.complex_branch4 = nn.Sequential(
CMaxPool2d(kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
complex_conv_block(in_channels, out_1x1pool, kernel_size=(1, 1)),
)
def forward(self, x):
return torch.cat(
[self.complex_branch1(x), self.complex_branch2(x), self.complex_branch3(x), self.complex_branch4(x)], 1
)
class complex_InceptionAux(nn.Module):
def __init__(self, in_channels, num_classes):
super(complex_InceptionAux, self).__init__()
self.relu = nn.ReLU()
self.dropout = nn.Dropout(p=0.7)
self.complex_pool = CAvgPool2d(kernel_size=5, stride=3)
self.complex_conv = complex_conv_block(in_channels, 128, kernel_size=1)
self.complex_linear1 = CLinear(2048, 1024)
self.complex_linear2 = CLinear(1024, num_classes)
def forward(self, x):
x = self.complex_pool(x)
x = self.complex_conv(x)
x = x.reshape(x.shape[0], -1, 2)
x = self.relu(self.complex_linear1(x))
x = self.dropout(x)
x = self.complex_linear2(x)
return x
class complex_conv_block(nn.Module):
def __init__(self, in_channels, out_channels, **kwargs):
super(complex_conv_block, self).__init__()
self.relu = nn.ReLU()
self.complex_conv = CConv2d(in_channels, out_channels, **kwargs)
self.complex_batchnorm = CBatchnorm(out_channels)
def forward(self, x):
return self.relu(self.complex_batchnorm(self.complex_conv(x)))