-
Notifications
You must be signed in to change notification settings - Fork 0
/
deep_caps_w_net.py
143 lines (113 loc) · 7.38 KB
/
deep_caps_w_net.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
from capsule_layers import *
import tensorflow as tf
import numpy as np
import cv2
def cap_U_encoder(input, K):
x = tf.reshape(input, shape=[-1, 224, 224, 3])
conv1 = tf.layers.conv2d(x, 16, 5, 1, padding="same", kernel_initializer=tf.truncated_normal_initializer(stddev=0.01))
print("Conv1_Shape : ", conv1.get_shape())
conv1_expanded = tf.expand_dims(conv1, axis=3)
#print("Conv1_Expanded Shape : ", conv1_expanded.get_shape())
conv_caps_1 = SegConvolution(conv1_expanded, 5, 2, 16, strides=2, routings=1)
print("Conv Caps 1 Shape : ", conv_caps_1.get_shape())
conv_caps_2 = SegConvolution(conv_caps_1, 5, 4, 16, strides=1, routings=3)
print("Conv Caps 2 Shape : ", conv_caps_2.get_shape())
conv_caps_3 = SegConvolution(conv_caps_2, 5, 4, 32, strides=2, routings=3)
print("Conv Caps 3 Shape : ", conv_caps_3.get_shape())
conv_caps_4 = SegConvolution(conv_caps_3, 5, 8, 32, strides=1, routings=3)
print("Conv Caps 4 Shape : ", conv_caps_4.get_shape())
conv_caps_5 = SegConvolution(conv_caps_4, 5, 8, 64, strides=2, routings=3)
print("Conv Caps 5 Shape : ", conv_caps_5.get_shape())
conv_caps_6 = SegConvolution(conv_caps_5, 5, 8, 64, strides=1, routings=3)
print("Conv Caps 6 Shape : ", conv_caps_6.get_shape())
conv_caps_7 = SegConvolution(conv_caps_6, 5, 8, 128, strides=2, routings=3)
print("Conv Caps 7 Shape : ", conv_caps_7.get_shape())
conv_caps_8 = SegConvolution(conv_caps_7, 5, 16, 64, strides=1, routings=3)
print("Conv Caps 8 Shape : ", conv_caps_8.get_shape())
deconv_caps_1 = SegConvolution(conv_caps_8, 4, 16, 64, strides=2, routings=3, op='deconv')
print("Deconv 1 Shape : ", deconv_caps_1.get_shape())
concat_1 = tf.concat([deconv_caps_1, conv_caps_6], axis=3)
conv_caps_9 = SegConvolution(concat_1, 5, 8, 32, strides=1, routings=3)
print("Conv Caps 9 Shape : ", conv_caps_9.get_shape())
deconv_caps_2 = SegConvolution(conv_caps_9, 4, 8, 32, strides=2, routings=3, op='deconv')
print("Deconv 2 Shape : ", deconv_caps_2.get_shape())
concat_2 = tf.concat([deconv_caps_2, conv_caps_4], axis=3)
conv_caps_10 = SegConvolution(concat_2, 5, 4, 16, strides=1, routings=3)
print("Conv Caps 10 Shape : ", conv_caps_10.get_shape())
deconv_caps_3 = SegConvolution(conv_caps_10, 4, 4, 16, strides=2, routings=3, op='deconv')
print("Deconv 2 Shape : ", deconv_caps_2.get_shape())
concat_3 = tf.concat([deconv_caps_3, conv_caps_2], axis=3)
conv_caps_11 = SegConvolution(concat_3, 5, 2, 16, strides=1, routings=3)
print("Conv Caps 11 Shape : ", conv_caps_11.get_shape())
deconv_caps_4 = SegConvolution(conv_caps_11, 4, 2, 16, strides=2, routings=3, op='deconv')
print("Deconv 4 Shape : ", deconv_caps_4.get_shape())
concat_4 = tf.concat([deconv_caps_4, conv1_expanded], axis=3)
conv_caps_12 = SegConvolution(concat_4, 1, 1, K, strides=1, routings=3)
print("Conv Caps 12 Shape : ", conv_caps_12.get_shape())
squeeze = tf.squeeze(conv_caps_12, axis=3)
#print("Squeeze Shape : ", squeeze.get_shape())
softmax = softmax_1 = tf.nn.softmax(squeeze)
print("Encode Output Shape : " , softmax.get_shape())
return softmax
def cap_U_decoder(input):
dec_conv1 = tf.layers.conv2d(input, 16, 5, 1, padding="same", kernel_initializer=tf.truncated_normal_initializer(stddev=0.01))
print("Conv1_Shape : ", dec_conv1.get_shape())
dec_conv1_expanded = tf.expand_dims(dec_conv1, axis=3)
#print("Conv1_Expanded Shape : ", conv1_expanded.get_shape())
dec_conv_caps_1 = SegConvolution(dec_conv1_expanded, 5, 2, 16, strides=2, routings=1)
print("Conv Caps 1 Shape : ", dec_conv_caps_1.get_shape())
dec_conv_caps_2 = SegConvolution(dec_conv_caps_1, 5, 4, 16, strides=1, routings=3)
print("Conv Caps 2 Shape : ", dec_conv_caps_2.get_shape())
dec_conv_caps_3 = SegConvolution(dec_conv_caps_2, 5, 4, 32, strides=2, routings=3)
print("Conv Caps 3 Shape : ", dec_conv_caps_3.get_shape())
dec_conv_caps_4 = SegConvolution(dec_conv_caps_3, 5, 8, 32, strides=1, routings=3)
print("Conv Caps 4 Shape : ", dec_conv_caps_4.get_shape())
dec_conv_caps_5 = SegConvolution(dec_conv_caps_4, 5, 8, 64, strides=2, routings=3)
print("Conv Caps 5 Shape : ", dec_conv_caps_5.get_shape())
dec_conv_caps_6 = SegConvolution(dec_conv_caps_5, 5, 8, 64, strides=1, routings=3)
print("Conv Caps 6 Shape : ", dec_conv_caps_6.get_shape())
dec_conv_caps_7 = SegConvolution(dec_conv_caps_6, 5, 8, 128, strides=2, routings=3)
print("Conv Caps 7 Shape : ", dec_conv_caps_7.get_shape())
dec_conv_caps_8 = SegConvolution(dec_conv_caps_7, 5, 16, 64, strides=1, routings=3)
print("Conv Caps 8 Shape : ", dec_conv_caps_8.get_shape())
dec_deconv_caps_1 = SegConvolution(dec_conv_caps_8, 4, 16, 64, strides=2, routings=3, op='deconv')
print("Deconv 1 Shape : ", dec_deconv_caps_1.get_shape())
dec_concat_1 = tf.concat([dec_deconv_caps_1, dec_conv_caps_6], axis=3)
dec_conv_caps_9 = SegConvolution(dec_concat_1, 5, 8, 32, strides=1, routings=3)
print("Conv Caps 9 Shape : ", dec_conv_caps_9.get_shape())
dec_deconv_caps_2 = SegConvolution(dec_conv_caps_9, 4, 8, 32, strides=2, routings=3, op='deconv')
print("Deconv 2 Shape : ", dec_deconv_caps_2.get_shape())
dec_concat_2 = tf.concat([dec_deconv_caps_2, dec_conv_caps_4], axis=3)
dec_conv_caps_10 = SegConvolution(dec_concat_2, 5, 4, 16, strides=1, routings=3)
print("Conv Caps 10 Shape : ", dec_conv_caps_10.get_shape())
dec_deconv_caps_3 = SegConvolution(dec_conv_caps_10, 4, 4, 16, strides=2, routings=3, op='deconv')
print("Deconv 3 Shape : ", dec_deconv_caps_3.get_shape())
dec_concat_3 = tf.concat([dec_deconv_caps_3, dec_conv_caps_2], axis=3)
dec_conv_caps_11 = SegConvolution(dec_concat_3, 5, 2, 16, strides=1, routings=3)
print("Conv Caps 11 Shape : ", dec_conv_caps_11.get_shape())
dec_deconv_caps_4 = SegConvolution(dec_conv_caps_11, 4, 2, 16, strides=2, routings=3, op='deconv')
print("Deconv 3 Shape : ", dec_deconv_caps_4.get_shape())
dec_concat_4 = tf.concat([dec_deconv_caps_4, dec_conv1_expanded], axis=3)
dec_conv_caps_12 = SegConvolution(dec_concat_4, 1, 1, 16, strides=1, routings=3)
print("Conv Caps 12 Shape : ", dec_conv_caps_12.get_shape())
dec_squeeze = tf.squeeze(dec_conv_caps_12, axis=3)
#Reconstruction
rec_conv_1 = tf.layers.conv2d(dec_squeeze, 64, 1, 1, padding="same", kernel_initializer=tf.truncated_normal_initializer(stddev=0.01))
rec_conv_2 = tf.layers.conv2d(rec_conv_1, 128, 1, 1, padding="same", kernel_initializer=tf.truncated_normal_initializer(stddev=0.01))
output = tf.layers.conv2d(rec_conv_2, 3, kernel_size = 1, padding="same")
print("Decoder Output Shape : " , output.get_shape())
return output
if __name__ == "__main__":
batch_size = 1
K = 4
width = 224
height = 224
img = cv2.imread("gdz.png")
img = cv2.resize(img, (width, height))
img = np.array([img])
input = np.random.uniform(size=(1, width, height, 3))
y = np.random.uniform(size=(1, width, height, 3))
image = tf.placeholder(tf.float32, [None, width, height, 3])
segment = tf.placeholder(tf.float32, [None, width, height, 3])
encoder = cap_U_encoder(image, K)
decoder = cap_U_decoder(encoder)