/
inception_v3.py
137 lines (116 loc) · 4 KB
/
inception_v3.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
from neupy.layers import *
from neupy import layers, plots
def ConvReluBN(*conv_args, **conv_kwargs):
return join(
Convolution(*conv_args, **conv_kwargs),
Relu(),
BatchNorm(epsilon=0.001),
)
def Inception_1(conv_filters):
return join(
parallel([
ConvReluBN((1, 1, conv_filters[0][0])),
], [
ConvReluBN((1, 1, conv_filters[1][0])),
ConvReluBN((5, 5, conv_filters[1][1]), padding=2),
], [
ConvReluBN((1, 1, conv_filters[2][0])),
ConvReluBN((3, 3, conv_filters[2][1]), padding=1),
ConvReluBN((3, 3, conv_filters[2][2]), padding=1),
], [
AveragePooling((3, 3), stride=(1, 1), padding='SAME'),
ConvReluBN((1, 1, conv_filters[3][0])),
]),
Concatenate(),
)
def Inception_2(conv_filters):
return join(
parallel([
ConvReluBN((1, 1, conv_filters[0][0])),
], [
ConvReluBN((1, 1, conv_filters[1][0])),
ConvReluBN((1, 7, conv_filters[1][1]), padding=(0, 3)),
ConvReluBN((7, 1, conv_filters[1][2]), padding=(3, 0)),
], [
ConvReluBN((1, 1, conv_filters[2][0])),
ConvReluBN((7, 1, conv_filters[2][1]), padding=(3, 0)),
ConvReluBN((1, 7, conv_filters[2][2]), padding=(0, 3)),
ConvReluBN((7, 1, conv_filters[2][3]), padding=(3, 0)),
ConvReluBN((1, 7, conv_filters[2][4]), padding=(0, 3)),
], [
AveragePooling((3, 3), stride=(1, 1), padding='SAME'),
ConvReluBN((1, 1, conv_filters[3][0])),
]),
Concatenate(),
)
def Inception_3(pooling):
pooling_layers = {'max': MaxPooling, 'avg': AveragePooling}
if pooling not in pooling_layers:
raise ValueError("Invalid pooling option: {}".format(pooling))
Pooling = pooling_layers[pooling]
return join(
parallel([
ConvReluBN((1, 1, 320)),
], [
ConvReluBN((1, 1, 384)),
parallel(
ConvReluBN((1, 3, 384), padding=(0, 1)),
ConvReluBN((3, 1, 384), padding=(1, 0)),
),
], [
ConvReluBN((1, 1, 448)),
ConvReluBN((3, 3, 384), padding=1),
parallel(
ConvReluBN((1, 3, 384), padding=(0, 1)),
ConvReluBN((3, 1, 384), padding=(1, 0)),
),
], [
Pooling((3, 3), stride=(1, 1), padding='SAME'),
ConvReluBN((1, 1, 192)),
]),
Concatenate(),
)
inception_v3 = join(
Input((299, 299, 3)),
ConvReluBN((3, 3, 32), stride=2),
ConvReluBN((3, 3, 32)),
ConvReluBN((3, 3, 64), padding=1),
MaxPooling((3, 3), stride=(2, 2)),
ConvReluBN((1, 1, 80)),
ConvReluBN((3, 3, 192)),
MaxPooling((3, 3), stride=(2, 2)),
Inception_1([[64], [48, 64], [64, 96, 96], [32]]),
Inception_1([[64], [48, 64], [64, 96, 96], [64]]),
Inception_1([[64], [48, 64], [64, 96, 96], [64]]),
parallel([
ConvReluBN((3, 3, 384), stride=2),
], [
ConvReluBN((1, 1, 64)),
ConvReluBN((3, 3, 96), padding=1),
ConvReluBN((3, 3, 96), stride=2),
], [
MaxPooling((3, 3), stride=(2, 2))
]),
Concatenate(),
Inception_2([[192], [128, 128, 192], [128, 128, 128, 128, 192], [192]]),
Inception_2([[192], [160, 160, 192], [160, 160, 160, 160, 192], [192]]),
Inception_2([[192], [160, 160, 192], [160, 160, 160, 160, 192], [192]]),
Inception_2([[192], [192, 192, 192], [192, 192, 192, 192, 192], [192]]),
parallel([
ConvReluBN((1, 1, 192)),
ConvReluBN((3, 3, 320), stride=2),
], [
ConvReluBN((1, 1, 192)),
ConvReluBN((1, 7, 192), padding=(0, 3)),
ConvReluBN((7, 1, 192), padding=(3, 0)),
ConvReluBN((3, 3, 192), stride=2),
], [
MaxPooling((3, 3), stride=(2, 2))
]),
Concatenate(),
Inception_3(pooling='avg'),
Inception_3(pooling='max'),
GlobalPooling('avg'),
Softmax(1000),
)
inception_v3.show()