-
Notifications
You must be signed in to change notification settings - Fork 12
Expand file tree
/
Copy pathmodel.py
More file actions
196 lines (149 loc) · 7.42 KB
/
model.py
File metadata and controls
196 lines (149 loc) · 7.42 KB
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
193
194
195
196
import tensorflow as tf
import numpy as np
def create_conv_3dlayer(input,
filter_width,
filter_height,
filter_depth,
stride=1,
num_output_channels=1,
relu=True):
layer = tf.layers.conv3d(inputs=input,
filters=num_output_channels,
kernel_size=[filter_width, filter_height, filter_depth],
strides=(1,1,stride),
padding='valid',
kernel_initializer=tf.truncated_normal_initializer(stddev=0.05),
bias_initializer=tf.constant_initializer(0.05),
data_format='channels_last')
if relu:
layer = tf.nn.relu(layer)
return layer
def create_conv_2dlayer(input,
filter_size,
num_output_channel,
relu=True,
pooling=False,
padding='valid',
d_format='channels_last'):
layer = tf.layers.conv2d(inputs=input, filters=num_output_channel,
kernel_size=[filter_size, filter_size],
padding=padding,
kernel_initializer=tf.truncated_normal_initializer(stddev=0.05),
bias_initializer=tf.constant_initializer(0.05),
data_format= d_format)
if pooling:
layer = tf.nn.max_pool(value=layer,
ksize=[1, 3, 3, 1],
strides=[1, 1, 1, 1],
padding='VALID')
if relu:
layer = tf.nn.relu(layer)
return layer
def create_conv_1dlayer(input,
filter_size,
num_output_channel,
stride=1,
relu=True,
padding='SAME'):
layer = tf.layers.conv1d(inputs=input, filters=num_output_channel, kernel_size=[filter_size], padding='valid',
kernel_initializer=tf.truncated_normal_initializer(stddev=0.05),
bias_initializer=tf.constant_initializer(0.05))
if relu:
layer = tf.nn.relu(layer)
return layer
def fully_connected_layer(input,
num_inputs,
num_outputs,
activation=None):
weights = tf.get_variable('weights', shape=[num_inputs, num_outputs])
biases = tf.get_variable('biases', shape=num_outputs)
layer = tf.matmul(input, weights) + biases
if activation is not None:
if activation == 'relu':
layer = tf.nn.relu(layer)
elif activation == 'softmax':
layer = tf.nn.softmax(layer)
return layer
def flatten_layer(layer):
layer_shape = layer.get_shape() # layer = [num_images, img_height, img_width, num_channels]
num_features = layer_shape[1:].num_elements() # Total number of elements in the network
layer_flat = tf.reshape(layer, [-1, num_features]) # -1 means total size of dimension is unchanged
return layer_flat, num_features
def net(statlieImg, prob, HEIGHT, WIDTH, CHANNELS, N_PARALLEL_BAND, NUM_CLASS):
sequence = {}
sequence['input'] = tf.reshape(statlieImg, [-1, HEIGHT, WIDTH, CHANNELS])
# Block 1 Conv1 layer
with tf.variable_scope('conv1'):
layer = sequence['input']
layer = create_conv_2dlayer(input=layer,
filter_size=1,
num_output_channel=CHANNELS,
relu=True,
padding='valid')
sequence['conv1'] = layer
# Tensor shape = N * 5 * 5 * 220
with tf.variable_scope('parallelProcess'):
layer = sequence['conv1']
with tf.variable_scope('reshape3d'):
layer = tf.reshape(layer, [-1, HEIGHT, WIDTH, CHANNELS, 1])
with tf.variable_scope('split'):
layer = tf.split(layer, num_or_size_splits=N_PARALLEL_BAND, axis=3)
with tf.variable_scope('segmentation', reuse=tf.AUTO_REUSE): # Enable parameter sharing
for index, l in enumerate(layer):
with tf.variable_scope('layer1'):
layer1 = create_conv_3dlayer(input=l,
filter_width=2,
filter_height=2,
filter_depth=9,
stride=2,
num_output_channels=1,
relu=True)
with tf.variable_scope('layer2'):
layer2 = create_conv_3dlayer(input=layer1,
filter_width=3,
filter_height=3,
filter_depth=5,
stride=1,
num_output_channels=3,
relu=True)
with tf.variable_scope('layer3'):
layer3 = create_conv_3dlayer(input=layer2,
filter_width=2,
filter_height=2,
filter_depth=3,
stride=2,
num_output_channels=6,
relu=True)
with tf.variable_scope('layer4'):
layer4 = create_conv_3dlayer(input=layer3,
filter_width=1,
filter_height=1,
filter_depth=3,
stride=1,
num_output_channels=10,
relu=True)
layer5, _ = flatten_layer(layer4)
if not index:
stack = tf.concat([layer5], axis=1)
else:
stack = tf.concat([stack, layer5], axis=1)
sequence['parallel_end'] = stack
with tf.variable_scope('dense1'):
layer = sequence['parallel_end']
layer, number_features = flatten_layer(layer)
layer = fully_connected_layer(input=layer,
num_inputs=number_features,
num_outputs=120,
activation='relu')
layer = tf.nn.dropout(x=layer, keep_prob=prob)
sequence['dense1'] = layer
with tf.variable_scope('dense3'):
layer = sequence['dense1']
layer = fully_connected_layer(input=layer,
num_inputs=120,
num_outputs=NUM_CLASS)
sequence['dense3'] = layer
y_predict = tf.nn.softmax(sequence['dense3'])
sequence['class_prediction'] = y_predict
sequence['predict_class_number'] = tf.argmax(y_predict, axis=1)
return sequence