-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathinception_3d.py
executable file
·193 lines (176 loc) · 9.49 KB
/
inception_3d.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
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
import numpy as np
import tensorflow as tf
from tensorflow.contrib.layers.python.layers import initializers
from tensorflow.contrib.layers.python.layers import regularizers
from tensorflow.contrib.framework.python.ops import arg_scope
def inception_module(input,scope,
num_b0,ks_b0,
num_b1_0,ks_b1_0,num_b1_1,ks_b1_1,
num_b2_0,ks_b2_0,num_b2_1,ks_b2_1,
ks_b3_0,num_b3_1,ks_b3_1):
end_point = scope
net = input
with tf.variable_scope(end_point):
with tf.variable_scope('Branch_0'):
branch_0 = tf.contrib.layers.conv3d(inputs=net, num_outputs=num_b0, kernel_size=ks_b0, scope='Conv3D_0a')
with tf.variable_scope('Branch_1'):
branch_1 = tf.contrib.layers.conv3d(inputs=net, num_outputs=num_b1_0, kernel_size=ks_b1_0, scope='Conv3D_0a')
branch_1 = tf.contrib.layers.conv3d(inputs=branch_1, num_outputs=num_b1_1, kernel_size=ks_b1_1, scope='Conv3D_0b')
with tf.variable_scope('Branch_2'):
branch_2 = tf.contrib.layers.conv3d(inputs=net, num_outputs=num_b2_0, kernel_size=ks_b2_0, scope='Conv3D_0a')
branch_2 = tf.contrib.layers.conv3d(inputs=branch_2, num_outputs=num_b2_1, kernel_size=ks_b2_1, scope='Conv3D_0b')
with tf.variable_scope('Branch_3'):
branch_3 = tf.contrib.layers.max_pool3d(inputs=net, kernel_size=ks_b3_0, stride=1, padding='same',scope='MaxPool3D_0a')
branch_3 = tf.contrib.layers.conv3d(inputs=branch_3, num_outputs=num_b3_1, kernel_size=ks_b3_1, scope='Conv3D_0b')
net = tf.concat([branch_0, branch_1, branch_2, branch_3], axis=4)
return net
def inception_3d_base(inputs,
final_endpoint='Inception_5b',
scope='Inception3D'):
end_points = {}
with tf.variable_scope(name_or_scope=scope,default_name="Inception3D",values=[inputs]):
with arg_scope([tf.contrib.layers.conv3d],
weights_initializer=initializers.xavier_initializer(),
weights_regularizer=regularizers.l2_regularizer(0.0002),
biases_initializer=tf.constant_initializer(0.2),
biases_regularizer=regularizers.l2_regularizer(0.0002)
):
with arg_scope([tf.contrib.layers.conv3d],
stride=1,
padding='SAME'):
end_point = 'Conv3D_1a'
net = tf.contrib.layers.conv3d(inputs=inputs,num_outputs=64,kernel_size=7,stride=2,scope='Conv3D_1a')
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'MaxPool3D_2a'
net = tf.contrib.layers.max_pool3d(inputs=net,kernel_size=3,stride=2,padding='same',scope='MaxPool3D_2a')
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Conv3D_2b'
net = tf.contrib.layers.conv3d(inputs=net,num_outputs=64,kernel_size=1,scope='Conv3D_2b')
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Conv3D_2c'
net = tf.contrib.layers.conv3d(inputs=net,num_outputs=192,kernel_size=3,scope='Conv3D_2c')
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'MaxPool3D_3a'
net = tf.contrib.layers.max_pool3d(inputs=net,kernel_size=3,stride=2,padding='same',scope='MaxPool3D_3a')
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_3a'
net = inception_module(net,end_point,
64,1,
96,1,128,3,
16,1,32,5,
3,32,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_3b'
net = inception_module(net,end_point,
128,1,
128,1,192,3,
32,1,96,5,
3,64,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'MaxPool3D_4a'
net = tf.contrib.layers.max_pool3d(inputs=net,kernel_size=3,stride=2,padding='same',scope=end_point)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_4a'
net = inception_module(net,end_point,
192,1,
96,1,208,3,
16,1,48,3,
3,64,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_4b'
net = inception_module(net,end_point,
160,1,
112,1,224,3,
24,1,64,3,
3,64,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_4c'
net = inception_module(net,end_point,
128,1,
128,1,256,3,
24,1,64,3,
3,64,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_4d'
net = inception_module(net,end_point,
112,1,
144,1,288,3,
32,1,64,3,
3,64,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_4e'
net = inception_module(net,end_point,
256,1,
160,1,320,3,
32,1,128,3,
3,128,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net,end_points
end_point = 'MaxPool3D_5a'
net = tf.contrib.layers.max_pool3d(inputs=net, kernel_size=3, stride=2,padding='same', scope=end_point)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_5a'
net = inception_module(net,end_point,
256,1,
160,1,320,3,
32,1,128,3,
3,128,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
end_point = 'Inception_5b'
net = inception_module(net,end_point,
384,1,
192,1,384,3,
48,1,128,3,
3,128,1)
end_points[end_point] = net
if final_endpoint == end_point:
return net, end_points
raise ValueError('Unknown final endpoint {0}'.format(final_endpoint))
def inception_3d_v1(inputs,
num_outputs=30,
dropout_keep_prob=0.8,
scope='Inception3D'):
with tf.variable_scope(scope, 'InceptionV1', [inputs]) as scope:
net, end_points = inception_3d_base(inputs=inputs,final_endpoint='Inception_5b')
with tf.variable_scope('Logits'):
net = tf.contrib.layers.max_pool3d(inputs=net,kernel_size=[1,7,7],stride=1,scope='AvgPool3D')
end_points['AvgPool3D'] = net
net = tf.layers.dropout(inputs=net,rate=dropout_keep_prob)
logits = tf.contrib.layers.fully_connected(inputs=net,num_outputs=num_outputs,activation_fn=None,
weights_initializer=tf.contrib.layers.xavier_initializer(),
biases_initializer=tf.constant_initializer(0.0),
weights_regularizer=regularizers.l2_regularizer(0.0002),
biases_regularizer=regularizers.l2_regularizer(0.0002),
scope='InnerProduct')
logits = tf.squeeze(net,[1,2,3],name='SpatialTemporalSqueeze')
end_points['Logits'] = logits
return logits, end_points