/
model_vgg_regression.py
178 lines (122 loc) · 6.26 KB
/
model_vgg_regression.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
import tensorflow as tf
import scipy
def weight_variable_accel(shape):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
def bias_variable_accel(shape):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial)
def conv2d_accel(x, W, stride):
return tf.nn.conv2d(x, W, strides=[1, stride, stride, 1], padding='SAME')
x_accel = tf.placeholder(tf.float32, shape=[None, 112, 112, 3])
y_accel_ = tf.placeholder(tf.float32, shape=[None, 1])
# isTheta = tf.placeholder(tf.bool)
x_image_accel = x_accel
# Block 1
#first convolutional layer
W_conv1_accel = weight_variable_accel([3, 3, 3, 64])
b_conv1_accel = bias_variable_accel([64])
h_conv1_accel = tf.nn.relu(conv2d_accel(x_image_accel, W_conv1_accel, 1) + b_conv1_accel)
#second convolutional layer
W_conv2_accel = weight_variable_accel([3, 3, 64, 64])
b_conv2_accel = bias_variable_accel([64])
h_conv2_accel = tf.nn.relu(conv2d_accel(h_conv1_accel, W_conv2_accel, 1) + b_conv2_accel)
h_conv2_accel_pool = tf.nn.max_pool(h_conv2_accel, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1],padding='SAME')
# Block 2
#third convolutional layer
W_conv3_accel = weight_variable_accel([3, 3, 64, 128])
b_conv3_accel = bias_variable_accel([128])
h_conv3_accel = tf.nn.relu(conv2d_accel(h_conv2_accel_pool, W_conv3_accel, 1) + b_conv3_accel)
#fourth convolutional layer
W_conv4_accel = weight_variable_accel([3, 3, 128, 128])
b_conv4_accel = bias_variable_accel([128])
h_conv4_accel = tf.nn.relu(conv2d_accel(h_conv3_accel, W_conv4_accel, 1) + b_conv4_accel)
h_conv4_accel_pool = tf.nn.max_pool(h_conv4_accel, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1],padding='SAME')
# Block 3
#fifth convolutional layer
W_conv5_accel = weight_variable_accel([3, 3, 128, 256])
b_conv5_accel = bias_variable_accel([256])
h_conv5_accel = tf.nn.relu(conv2d_accel(h_conv4_accel_pool, W_conv5_accel, 1) + b_conv5_accel)
#sixth convolutional layer
W_conv6_accel = weight_variable_accel([3, 3, 256, 256])
b_conv6_accel = bias_variable_accel([256])
h_conv6_accel = tf.nn.relu(conv2d_accel(h_conv5_accel, W_conv6_accel, 1) + b_conv6_accel)
#seventh convolutional layer
W_conv7_accel = weight_variable_accel([3, 3, 256, 256])
b_conv7_accel = bias_variable_accel([256])
h_conv7_accel = tf.nn.relu(conv2d_accel(h_conv6_accel, W_conv7_accel, 1) + b_conv7_accel)
#eigth convolutional layer
W_conv8_accel = weight_variable_accel([3, 3, 256, 256])
b_conv8_accel = bias_variable_accel([256])
h_conv8_accel = tf.nn.relu(conv2d_accel(h_conv7_accel, W_conv8_accel, 1) + b_conv8_accel)
h_conv8_accel_pool = tf.nn.max_pool(h_conv8_accel, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1],padding='SAME')
# Block 4
#9th convolutional layer
W_conv9_accel = weight_variable_accel([3, 3, 256, 512])
b_conv9_accel = bias_variable_accel([512])
h_conv9_accel = tf.nn.relu(conv2d_accel(h_conv8_accel_pool, W_conv9_accel, 1) + b_conv9_accel)
#10th convolutional layer
W_conv10_accel = weight_variable_accel([3, 3, 512, 512])
b_conv10_accel = bias_variable_accel([512])
h_conv10_accel = tf.nn.relu(conv2d_accel(h_conv9_accel, W_conv10_accel, 1) + b_conv10_accel)
#11th convolutional layer
W_conv11_accel = weight_variable_accel([3, 3, 512, 512])
b_conv11_accel = bias_variable_accel([512])
h_conv11_accel = tf.nn.relu(conv2d_accel(h_conv10_accel, W_conv11_accel, 1) + b_conv11_accel)
#12th convolutional layer
W_conv12_accel = weight_variable_accel([3, 3, 512, 512])
b_conv12_accel = bias_variable_accel([512])
h_conv12_accel = tf.nn.relu(conv2d_accel(h_conv11_accel, W_conv12_accel, 1) + b_conv12_accel)
h_conv12_accel_pool = tf.nn.max_pool(h_conv12_accel, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1],padding='SAME')
# Block 5
#13th convolutional layer
W_conv13_accel = weight_variable_accel([3, 3, 512, 512])
b_conv13_accel = bias_variable_accel([512])
h_conv13_accel = tf.nn.relu(conv2d_accel(h_conv12_accel_pool, W_conv13_accel, 1) + b_conv13_accel)
#14th convolutional layer
W_conv14_accel = weight_variable_accel([3, 3, 512, 512])
b_conv14_accel = bias_variable_accel([512])
h_conv14_accel = tf.nn.relu(conv2d_accel(h_conv13_accel, W_conv14_accel, 1) + b_conv14_accel)
#15th convolutional layer
W_conv15_accel = weight_variable_accel([3, 3, 512, 512])
b_conv15_accel = bias_variable_accel([512])
h_conv15_accel = tf.nn.relu(conv2d_accel(h_conv14_accel, W_conv15_accel, 1) + b_conv15_accel)
#16th convolutional layer
W_conv16_accel = weight_variable_accel([3, 3, 512, 512])
b_conv16_accel = bias_variable_accel([512])
h_conv16_accel = tf.nn.relu(conv2d_accel(h_conv15_accel, W_conv16_accel, 1) + b_conv16_accel)
h_conv16_accel_pool = tf.nn.max_pool(h_conv16_accel, ksize=[1, 2, 2, 1],strides=[1, 2, 2, 1],padding='SAME')
#FCL 1
W_fc1_accel = weight_variable_accel([16*512, 512])
b_fc1_accel = bias_variable_accel([512])
# print(h_conv4_accel_pool.shape)
h_conv16_flat_accel = tf.reshape(h_conv16_accel_pool, [-1, 16*512])
h_fc1_accel = tf.nn.relu(tf.matmul(h_conv16_flat_accel, W_fc1_accel) + b_fc1_accel)
keep_prob_accel = tf.placeholder(tf.float32)
h_fc1_drop_accel = tf.nn.dropout(h_fc1_accel, keep_prob_accel)
#FCL 2
W_fc2_accel = weight_variable_accel([512, 1])
b_fc2_accel = bias_variable_accel([1])
y_accel = tf.multiply(tf.atan(tf.matmul(h_fc1_drop_accel, W_fc2_accel) + b_fc2_accel), 2)
# y_accel = tf.matmul(h_fc1_drop_accel, W_fc2_accel) + b_fc2_accel
# h_fc2_accel = tf.nn.relu(tf.matmul(h_fc1_drop_accel, W_fc2_accel) + b_fc2_accel)
# h_fc2_drop_accel = tf.nn.dropout(h_fc2_accel, keep_prob_accel)
# #FCL 3
# W_fc3_accel = weight_variable_accel([100, 50])
# b_fc3_accel = bias_variable_accel([50])
# h_fc3_accel = tf.nn.relu(tf.matmul(h_fc2_drop_accel, W_fc3_accel) + b_fc3_accel)
# h_fc3_drop_accel = tf.nn.dropout(h_fc3_accel, keep_prob_accel)
# #FCL 3
# W_fc4_accel = weight_variable_accel([50, 10])
# b_fc4_accel = bias_variable_accel([10])
# h_fc4_accel = tf.nn.relu(tf.matmul(h_fc3_drop_accel, W_fc4_accel) + b_fc4_accel)
# h_fc4_drop_accel = tf.nn.dropout(h_fc4_accel, keep_prob_accel)
# #Output
# W_fc5_accel = weight_variable_accel([10, 1])
# b_fc5_accel = bias_variable_accel([1])
# # if (tf.cond(tf.eisTheta = True)):
# # if (tf.cond(tf.equal(isTheta, tf.constant(True)), lambda: True, lambda: False)):
# # if (isTheta is not None):
# y_accel = tf.multiply(tf.atan(tf.matmul(h_fc4_drop_accel, W_fc5_accel) + b_fc5_accel), 2) #scale the atan output
# else:
# y = tf.matmul(h_fc4_drop, W_fc5) + b_fc5 #linear output