-
Notifications
You must be signed in to change notification settings - Fork 1
/
model.py
150 lines (131 loc) · 6.54 KB
/
model.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
# -*- coding: utf-8 -*-
"""
-------------------------------------------------
File Name： model
Description :
Author : iffly
date： 4/23/18
-------------------------------------------------
Change Activity:
4/23/18:
-------------------------------------------------
"""
import tensorflow as tf
class C3d(object):
def __init__(self, num_class=20, keep_prob=0.5, wd=0.00005, frame_num=16, size_w=112, size_h=112, chanel_num=3,
weight_init=tf.contrib.layers.xavier_initializer()):
self.num_class = num_class
self.keep_prob = keep_prob
self.frame_num = frame_num
self.size_w = size_w
self.size_h = size_h
self.chanel_num = chanel_num
self.wd = wd
self.weight_init = weight_init
def _variable(self, name, shape, initializer):
with tf.device("/cpu:0"):
var = tf.get_variable(name, shape, initializer=initializer)
return var
def _variable_with_weight_decay(self, name, shape, stddev, wd):
var = self._variable(name, shape, self.weight_init)
if wd is not None:
weight_decay = tf.multiply(tf.nn.l2_loss(var), wd)
tf.add_to_collection('weight_decay_loss', weight_decay)
return var
def conv3d(self, name, l_input, w, b):
with tf.variable_scope(name) as scope:
return tf.nn.bias_add(
tf.nn.conv3d(l_input, w, strides=[1, 1, 1, 1, 1], padding='SAME'),
b
)
def max_pool(self, name, l_input, k):
return tf.nn.max_pool3d(l_input, ksize=[1, k, 2, 2, 1], strides=[1, k, 2, 2, 1], padding='SAME', name=name)
def build_weight(self):
with tf.variable_scope('var_name') as var_scope:
wd = self.wd
self.weights = {
'wc1': self._variable_with_weight_decay('wc1', [3, 3, 3, 3, 64], 0.04, wd),
'wc2': self._variable_with_weight_decay('wc2', [3, 3, 3, 64, 128], 0.04, wd),
'wc3a': self._variable_with_weight_decay('wc3a', [3, 3, 3, 128, 256], 0.04, wd),
'wc3b': self._variable_with_weight_decay('wc3b', [3, 3, 3, 256, 256], 0.04, wd),
'wc4a': self._variable_with_weight_decay('wc4a', [3, 3, 3, 256, 512], 0.04, wd),
'wc4b': self._variable_with_weight_decay('wc4b', [3, 3, 3, 512, 512], 0.04, wd),
'wc5a': self._variable_with_weight_decay('wc5a', [3, 3, 3, 512, 512], 0.04, wd),
'wc5b': self._variable_with_weight_decay('wc5b', [3, 3, 3, 512, 512], 0.04, wd),
'wd1': self._variable_with_weight_decay('wd1', [8192, 4096], 0.04, wd),
'wd2': self._variable_with_weight_decay('wd2', [4096, 4096], 0.04, wd),
'out': self._variable_with_weight_decay('wout', [4096, self.num_class], 0.04, wd)
}
self.biases = {
'bc1': self._variable_with_weight_decay('bc1', [64], 0.04, None),
'bc2': self._variable_with_weight_decay('bc2', [128], 0.04, None),
'bc3a': self._variable_with_weight_decay('bc3a', [256], 0.04, None),
'bc3b': self._variable_with_weight_decay('bc3b', [256], 0.04, None),
'bc4a': self._variable_with_weight_decay('bc4a', [512], 0.04, None),
'bc4b': self._variable_with_weight_decay('bc4b', [512], 0.04, None),
'bc5a': self._variable_with_weight_decay('bc5a', [512], 0.04, None),
'bc5b': self._variable_with_weight_decay('bc5b', [512], 0.04, None),
'bd1': self._variable_with_weight_decay('bd1', [4096], 0.04, None),
'bd2': self._variable_with_weight_decay('bd2', [4096], 0.04, None),
'out': self._variable_with_weight_decay('bout', [self.num_class], 0.04, None),
}
def getweights(self):
return self.weights
def getbiases(self):
return self.biases
def build_model(self, input=None, weights=None):
if weights:
self.weights = weights
else:
self.build_weight()
if not input is None:
self.input = input
else:
self.input = tf.placeholder(tf.float32, shape=(None,
self.frame_num,
self.size_w,
self.size_w,
self.chanel_num))
_weights = self.weights
_biases = self.biases
conv3d = self.conv3d
max_pool = self.max_pool
_dropout = self.keep_prob
# Convolution Layer
net = conv3d('conv1', self.input, _weights['wc1'], _biases['bc1'])
net = tf.nn.relu(net, 'relu1')
net = max_pool('pool1', net, k=1)
# Convolution Layer
net = conv3d('conv2', net, _weights['wc2'], _biases['bc2'])
net = tf.nn.relu(net, 'relu2')
net = max_pool('pool2', net, k=2)
# Convolution Layer
net = conv3d('conv3a', net, _weights['wc3a'], _biases['bc3a'])
net = tf.nn.relu(net, 'relu3a')
net = conv3d('conv3b', net, _weights['wc3b'], _biases['bc3b'])
net = tf.nn.relu(net, 'relu3b')
net = max_pool('pool3', net, k=2)
# Convolution Layer
net = conv3d('conv4a', net, _weights['wc4a'], _biases['bc4a'])
net = tf.nn.relu(net, 'relu4a')
net = conv3d('conv4b', net, _weights['wc4b'], _biases['bc4b'])
net = tf.nn.relu(net, 'relu4b')
net = max_pool('pool4', net, k=2)
# Convolution Layer
net = conv3d('conv5a', net, _weights['wc5a'], _biases['bc5a'])
net = tf.nn.relu(net, 'relu5a')
net = conv3d('conv5b', net, _weights['wc5b'], _biases['bc5b'])
net = tf.nn.relu(net, 'relu5b')
net = max_pool('pool5', net, k=2)
# Fully connected layer
net = tf.transpose(net, perm=[0, 1, 4, 2, 3])
net = tf.reshape(net, [-1, _weights['wd1'].get_shape().as_list()[
0]]) # Reshape conv3 output to fit dense layer input
net = tf.nn.bias_add(tf.matmul(net, _weights['wd1']), _biases['bd1'])
net = tf.nn.relu(net, name='fc1') # Relu activation
net = tf.nn.dropout(net, _dropout)
net = tf.nn.relu(tf.nn.bias_add(tf.matmul(net, _weights['wd2']), _biases['bd2']), name='fc2') # Relu activation
net = tf.nn.dropout(net, _dropout)
# Output: class prediction
net = tf.nn.bias_add(tf.matmul(net, _weights['out']), _biases['out'], name='out')
return net