-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathres_unet.py
90 lines (69 loc) · 3.62 KB
/
res_unet.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
"""
the author is leilei
"""
'''
链接到结构图片：
https://github.com/ogvalt/deep_residual_unet/blob/master/architecture.jpg?raw=true
'''
import tensorflow as tf
from tensorflow.contrib import slim
from tensorflow.contrib.layers import xavier_initializer
class Res_Unet:
def __init__(self, class_number):
self.class_number = class_number
def upsampling(self, bottom, feature_map_size):
# feature_map_size: int [h,w]
return tf.image.resize_bilinear(bottom, size=feature_map_size)
def res_block(self, bottom, num_outputs, strides, is_training):
'''
also: slim.batch_norm(activation_fn=slim.nn.relu)
'''
out = slim.batch_norm(bottom, center=True, scale=True, is_training=is_training)
out = slim.nn.relu(out)
out = slim.conv2d(out, num_outputs[0], 3, strides[0], padding='SAME', activation_fn=None)
out = slim.batch_norm(out, center=True, scale=True, is_training=is_training)
out = slim.nn.relu(out)
out = slim.conv2d(out, num_outputs[1], 3, strides[1], padding='SAME', activation_fn=None)
shortcut = slim.conv2d(bottom, num_outputs[1], 1, strides[0], padding='SAME', activation_fn=None)
shortcut = slim.batch_norm(shortcut, center=True, scale=True, is_training=is_training)
output = tf.add(shortcut, out)
return output
def encoder(self, bottom, is_training):
to_decoder = []
out = slim.conv2d(bottom, 64, 3, 1, padding='SAME', activation_fn=None)
out = slim.batch_norm(out, center=True, scale=True, is_training=is_training)
out = slim.nn.relu(out)
out = slim.conv2d(out, 64, 3, 1, padding='SAME', activation_fn=None)
shortcut = slim.conv2d(bottom, 64, 1, 1, padding='SAME', activation_fn=None)
shortcut = slim.batch_norm(shortcut, center=True, scale=True, is_training=is_training)
output = tf.add(shortcut, out)
to_decoder.append(output)
output = self.res_block(output, num_outputs=[128, 128], strides=[2, 1], is_training)
to_decoder.append(output)
output = self.res_block(output, num_outputs=[256, 256], strides=[2, 1], is_training)
to_decoder.append(output)
return to_decoder
def decoder(self, bottom, from_encoder, is_training):
out = self.upsampling(bottom, tf.shape(from_encoder[2])[1:3])
out = tf.concat([out, from_encoder[2]], axis=-1)
out = self.res_block(out, num_outputs=[256, 256], strides=[1, 1], is_training)
out = self.upsampling(out, tf.shape(from_encoder[1])[1:3])
out = tf.concat([out, from_encoder[1]], axis=-1)
out = self.res_block(out, num_outputs=[128, 128], strides=[1, 1], is_training)
out = self.upsampling(out, tf.shape(from_encoder[0])[1:3])
out = tf.concat([out, from_encoder[0]], axis=-1)
output = self.res_block(out, num_outputs=[64, 64], strides=[1, 1], is_training)
return output
def build(self, image, is_training):
with tf.name_scope('processing'):
b, g, r = tf.split(image, 3, axis=3)
image = tf.concat([
b * 0.00390625,
g * 0.00390625,
r * 0.00390625], axis=3)
self.to_decoder = self.encoder(image, is_training)
self.middle = self.res_block(self.to_decoder[2], num_outputs=[512, 512], strides=[2, 1], is_training)
self.output = self.decoder(self.middle, self.to_decoder, is_training)
self.score = slim.conv2d(self.output, self.class_number, 1, 1, activation_fn=None)
self.softmax = slim.nn.softmax(self.score + tf.constant(1e-4))
self.pred = tf.argmax(self.softmax, axis=-1)