/
models.py
executable file
·209 lines (170 loc) · 6.7 KB
/
models.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
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
#!/usr/bin/env python
import numpy as np
import keras
from keras.models import Model
from keras.layers import Input, concatenate, GaussianNoise, Flatten, Dense, Reshape, BatchNormalization, Dropout
from keras.layers import Conv2D, MaxPooling2D, Conv2DTranspose , LeakyReLU, Activation, Add, Concatenate
from keras.optimizers import Adam
from keras.callbacks import ModelCheckpoint
from keras import backend as K
from keras.regularizers import l2
import json, glob, random, sys
import losses
K.set_image_data_format('channels_last') # TF dimension ordering in this code
def getModel(name, shape):
if name == "Identity":
return get_ident(shape)
if name == "FC":
return get_fc(shape)
if "-" in name:
mod = name.split("-")[0]
fil = int(name.split("-")[1][0])
if mod == "DilCN":
return get_dilcn(shape, fil)
if mod == "FullCN":
return get_fullcn(shape, fil)
if mod == "FullBN":
return get_fullbn(shape, fil)
if mod == "Residual":
return get_resnet(shape, fil)
if mod == "CascadeNet":
return get_cascadenet(shape, fil)
print "Unknown model: %s " % name
sys.exit()
def get_ident((img_rows, img_cols, channels )):
inputs = Input((img_rows, img_cols, channels))
output = inputs
model = Model(inputs=[inputs], outputs=[output])
model.compile(optimizer=Adam(lr=1e-3), loss=losses.mae)
return model
def get_fc((img_rows, img_cols, channels )):
inputs = Input((img_rows, img_cols, channels))
noise = GaussianNoise(0.5)(inputs)
flat = Flatten()(noise)
dense = Dense(512, activation='relu')(flat)
dense = Dense(512, activation='relu')(dense)
dense = Dense(img_rows * img_cols)(dense)
output = Reshape((img_rows, img_cols, 1))(dense)
model = Model(inputs=[inputs], outputs=[output])
model.compile(optimizer=Adam(lr=1e-3), loss=losses.mae)
return model
def get_dilcn((img_rows, img_cols, channels ), num_cn):
inputs = Input((img_rows, img_cols, channels))
conv1 = GaussianNoise(0.5)(inputs)
#conv1 = inputs
for i in range(num_cn):
conv1 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv1)
#conv1 = LeakyReLU()(conv1)
#conv1 = BatchNormalization()(conv1)
conv2 = Conv2D(64, (3, 3), dilation_rate=2, padding='same')(conv1)
for i in range(num_cn):
conv2 = Conv2D(128, (3, 3), activation='relu',padding='same')(conv2)
#conv2 = LeakyReLU()(conv2)
#conv2 = BatchNormalization()(conv2)
conv3 = Conv2D(128, (3, 3), dilation_rate=2, padding='same')(conv2)
for i in range(num_cn):
conv3 = Conv2D(256, (3, 3), activation='relu',padding='same')(conv3)
#conv3 = LeakyReLU()(conv3)
#conv3 = BatchNormalization()(conv3)
conv4 = Conv2D(256, (3, 3), dilation_rate=2, padding='same')(conv3)
conv4 = Conv2D(1, (1, 1))( conv4)
model = Model(inputs=[inputs], outputs=[conv4])
model.compile(optimizer=Adam(lr=1e-5), loss=losses.mae)
return model
def cascade_block(x, filters, size):
x = BatchNormalization(axis=-1)(x)
x = Activation("relu")(x)
x = Conv2D(filters, (size, size), padding='same', kernel_initializer="he_normal")(x)
return x
def get_cascadenet((img_rows, img_cols, channels ), num_cn):
input = Input((img_rows, img_cols, channels))
shortcuts = [input]
x = input
for i in range(num_cn):
x = cascade_block(x, 64, 3)
shortcuts.append(x)
x = Concatenate()(shortcuts)
for i in range(num_cn):
x = cascade_block(x, 128, 3)
shortcuts.append(x)
x = Concatenate()(shortcuts)
for i in range(num_cn):
x = cascade_block(x, 256, 3)
shortcuts.append(x)
x = Concatenate()(shortcuts)
conv4 = cascade_block(x, 1, 1)
model = Model(inputs=[input], outputs=[conv4])
model.compile(optimizer=Adam(lr=1e-3), loss=losses.mae)
return model
def res_block(x, filters, size, last=False):
shortcut = x
x = BatchNormalization(axis=-1)(x)
x = Activation("relu")(x)
x = Conv2D(filters, (size, size), padding='same', kernel_initializer="he_normal")(x)
if last:
return x
return Add()([shortcut, x])
def get_resnet((img_rows, img_cols, channels ), num_cn):
inputs = Input((img_rows, img_cols, channels))
conv1 = inputs
#conv1 = GaussianNoise(0.5)(inputs)
#conv1 = inputs
for i in range(num_cn):
conv1 = res_block(conv1, 64, 3)
conv2 = Conv2D(128, (1, 1), padding="same", kernel_initializer="he_normal")(conv1)
for i in range(num_cn):
conv2 = res_block(conv2, 128, 3)
conv3 = Conv2D(256, (1, 1), padding="same", kernel_initializer="he_normal")(conv2)
for i in range(num_cn):
conv3 = res_block(conv3, 256, 3)
conv4 = res_block(conv3, 1, 1, True)
model = Model(inputs=[inputs], outputs=[conv4])
model.compile(optimizer=Adam(lr=1e-3), loss=losses.mae)
return model
def get_fullbn((img_rows, img_cols, channels ), num_cn):
inputs = Input((img_rows, img_cols, channels))
conv1 = inputs
conv1 = BatchNormalization(axis=-1)(conv1)
#conv1 = GaussianNoise(0.5)(inputs)
#conv1 = inputs
for i in range(num_cn):
conv1 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv1)
#conv1 = LeakyReLU()(conv1)
conv1 = BatchNormalization(axis=-1)(conv1)
conv2 = conv1
for i in range(num_cn):
conv2 = Conv2D(128, (3, 3), activation='relu',padding='same')(conv2)
#conv2 = LeakyReLU()(conv2)
conv2 = BatchNormalization(axis=-1)(conv2)
conv3 = conv2
for i in range(num_cn):
conv3 = Conv2D(256, (3, 3), activation='relu',padding='same')(conv3)
#conv3 = LeakyReLU()(conv3)
conv3 = BatchNormalization(axis=-1)(conv3)
conv4 = Conv2D(1, (1, 1))( conv3)
model = Model(inputs=[inputs], outputs=[conv4])
model.compile(optimizer=Adam(lr=1e-3), loss=losses.mae)
return model
def get_fullcn((img_rows, img_cols, channels ), num_cn):
inputs = Input((img_rows, img_cols, channels))
conv1 = inputs
#conv1 = GaussianNoise(0.5)(inputs)
#conv1 = inputs
for i in range(num_cn):
conv1 = Conv2D(64, (3, 3), activation='relu', padding='same')(conv1)
#conv1 = LeakyReLU()(conv1)
#conv1 = BatchNormalization()(conv1)
conv2 = conv1
for i in range(num_cn):
conv2 = Conv2D(128, (3, 3), activation='relu',padding='same')(conv2)
#conv2 = LeakyReLU()(conv2)
#conv2 = BatchNormalization()(conv2)
conv3 = conv2
for i in range(num_cn):
conv3 = Conv2D(256, (3, 3), activation='relu',padding='same')(conv3)
#conv3 = LeakyReLU()(conv3)
#conv3 = BatchNormalization()(conv3)
conv4 = Conv2D(1, (1, 1))( conv3)
model = Model(inputs=[inputs], outputs=[conv4])
model.compile(optimizer=Adam(lr=1e-5), loss=losses.mae)
return model