-
Notifications
You must be signed in to change notification settings - Fork 70
/
models.py
executable file
·173 lines (126 loc) · 5.81 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
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Dense, Activation, Flatten, Reshape
from tensorflow.keras.layers import Conv2D, Conv2DTranspose
from tensorflow.keras.layers import LeakyReLU, Dropout
from tensorflow.keras import initializers
from tensorflow.keras import backend as K
from tensorflow.keras.optimizers import Adam
def set_trainability(model, trainable=False):
model.trainable = trainable
for layer in model.layers:
layer.trainable = trainable
# LSGAN Model
class LSGAN_Model(object):
def __init__(self, img_rows=28, img_cols=28, channel=1, noise_dim=100, dataset='MNIST'):
self.dataset = dataset
self.img_rows = img_rows
self.img_cols = img_cols
self.channel = channel
self.noise_dim = noise_dim
self.D = None # discriminator
self.G = None # generator
self.AM = None # adversarial model
self.DM = None
self.optimizer = Adam(lr=0.00005, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
def discriminator(self):
if self.D:
return self.D
# kern_init = initializers.RandomNormal(mean=0.0, stddev=0.02, seed=None)
kern_init = initializers.glorot_normal()
input_shape = (self.img_rows, self.img_cols, self.channel)
input_img = Input(shape=input_shape, name='Input_Image')
x = Conv2D(16, 5, strides=2, input_shape=input_shape, padding='same', kernel_initializer=kern_init)(input_img)
# x = BatchNormalization()(x)
x = LeakyReLU(alpha=0.2)(x)
x = Conv2D(32, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = LeakyReLU(alpha=0.2)(x)
x = Conv2D(64, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = LeakyReLU(alpha=0.2)(x)
x = Conv2D(128, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = LeakyReLU(alpha=0.2)(x)
x = Conv2D(256, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = LeakyReLU(alpha=0.2)(x)
x = Conv2D(512, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = LeakyReLU(alpha=0.2)(x)
# Out: 1-dim probability
x = Flatten()(x)
x = Dense(1, activation='sigmoid')(x)
self.D = Model(inputs=input_img, outputs=x, name='Discriminator')
self.D.summary()
return self.D
def generator(self):
if self.G:
return self.G
kern_init = initializers.glorot_normal()
input_shape = (self.noise_dim,)
input_noise = Input(shape=input_shape, name='noise')
dim = 7
depth = 512
x = Dense(dim * dim * depth, kernel_initializer=kern_init)(input_noise)
# x = BatchNormalization()(x)
# x = Activation('relu')(x)
x = Reshape((dim, dim, depth))(x)
x = Conv2DTranspose(depth / 2, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = Activation('selu')(x)
x = Conv2DTranspose(depth / 4, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = Activation('selu')(x)
x = Conv2DTranspose(depth / 8, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = Activation('selu')(x)
x = Conv2DTranspose(depth / 16, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
# x = BatchNormalization()(x)
x = Activation('selu')(x)
x = Conv2DTranspose(self.channel, 5, strides=2, padding='same', kernel_initializer=kern_init)(x)
x = Activation('tanh')(x)
self.G = Model(inputs=input_noise, outputs=x, name='Generator')
self.G.summary()
return self.G
def adversarial_model(self, gen, dis):
if self.AM:
return self.AM
input_shape = (self.noise_dim,)
input_noise_AM = Input(shape=input_shape, name='noise')
img_fake = gen(input_noise_AM)
x = dis(img_fake)
out_AM = Dropout(1.0, name='out_img_fake')(x)
self.AM = Model(inputs=input_noise_AM, outputs=out_AM)
set_trainability(dis, False)
self.AM.compile(loss=self.loss_LSGAN, optimizer=self.optimizer, metrics=['accuracy'])
self.AM.summary()
return self.AM
def discriminator_model(self, dis):
if self.DM:
return self.DM
input_shape = (self.img_rows, self.img_cols, self.channel)
input_img = Input(shape=input_shape)
x = dis(input_img)
self.DM = Model(inputs=input_img, outputs=x)
self.DM.compile(loss=self.loss_LSGAN, optimizer=self.optimizer, metrics=['accuracy'])
return self.DM
@staticmethod
def loss_LSGAN(y_true, y_pred):
return K.mean(K.square(y_pred-y_true), axis=-1)/2
# Completion Model
class Completion_Model(object):
def __init__(self, noise_dim=100):
self.noise_dim = noise_dim
self.CL = None
# complete loss = contextural loss + perceptual loss
def cal_complete_loss(self, gen, dis):
if self.CL:
return self.CL
input_shape = (self.noise_dim,)
input_noise_CL = Input(shape=input_shape, name='noise')
out_gen_img = gen(input_noise_CL)
out_gen_img = Dropout(1.0, name='name_out_gen_img')(out_gen_img)
out_dis_val = dis(out_gen_img)
out_dis_val = Dropout(1.0, name='name_out_dis_val')(out_dis_val)
self.CL = Model(inputs=input_noise_CL, outputs=[out_dis_val, out_gen_img])
return self.CL