Repository navigation
Expand file tree
/
Copy pathBaseModel.py
More file actions
127 lines (92 loc) · 3.37 KB
/
Copy pathBaseModel.py
File metadata and controls
127 lines (92 loc) · 3.37 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
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
'''
Shortcut Network Model
'''
from keras.models import Model
from keras.layers import Conv2D, BatchNormalization, Activation, Input
from keras.layers.merge import add
class ShortCutNet():
'''Generate networks with short cut
'''
def __init__(self):
'''Initialization
'''
self.num_filters = 128
self.kernel_size = (3, 3)
self.strides = (1, 1)
self.padding = "same"
self.kernel_initializer = "he_normal"
#self.kernel_regularizer = l2(1.e-4)
self.kernel_regularizer = None
self.trainable = True
def _BNRelu(self, x):
"""Helper to build a BN with relu block
"""
x = BatchNormalization()(x)
return Activation("relu")(x)
def _ConvOnly(self, x):
return Conv2D(filters = self.num_filters,
kernel_size = self.kernel_size,
strides = self.strides,
padding = self.padding,
kernel_initializer = self.kernel_initializer,
kernel_regularizer = self.kernel_regularizer,
trainable = self.trainable)(x)
def _ConvBNRelu(self, x):
x = Conv2D(filters = self.num_filters,
kernel_size = self.kernel_size,
strides = self.strides,
padding = self.padding,
kernel_initializer = self.kernel_initializer,
kernel_regularizer = self.kernel_regularizer,
trainable = self.trainable)(x)
return self._BNRelu(x)
def _AddShortCut(self, x):
x = add(x)
return self._BNRelu(x)
def DeblurResidualNet(self, input_shape, num_block):
'''Base residual net network for deblur.
'''
input_x = Input(shape=input_shape)
# Model
# trainable flag just for debug
self.trainable = True
#self.trainable = False
self.kernel_size = (25, 25)
self.padding = "valid"
layer_25by25 = self._ConvBNRelu(input_x)
self.kernel_size = (3, 3)
self.padding = "same"
shortcut = self._ConvBNRelu(layer_25by25)
for i in xrange(0, num_block):
x = self._ConvBNRelu(shortcut)
x = self._ConvBNRelu(x)
x = self._ConvOnly(x)
shortcut = self._AddShortCut([x, shortcut])
self.num_filters = 1
x = self._ConvOnly(shortcut)
model = Model(inputs=input_x, outputs=x)
return model
def DeblurSHCNet(self, input_shape, num_block):
'''Base SHC net network for deblur.
'''
input_x = Input(shape=input_shape)
# Model
# trainable flag just for debug
self.trainable = True
#self.trainable = False
self.kernel_size = (25, 25)
self.padding = "valid"
layer_25by25 = self._ConvBNRelu(input_x)
self.kernel_size = (3, 3)
self.padding = "same"
shortcut = self._ConvBNRelu(layer_25by25)
x = self._ConvBNRelu(shortcut)
for i in xrange(0, num_block):
x = self._ConvOnly(x)
x = self._AddShortCut([x, shortcut])
self.num_filters = 1
x = self._ConvOnly(x)
model = Model(inputs=input_x, outputs=x)
return model