-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmodels.py
170 lines (132 loc) · 7.13 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
from keras.models import Model
from keras.layers.core import Flatten, Dense, Activation, Dropout
from keras.layers.convolutional import Conv2D, UpSampling2D
from keras.layers import Input, Concatenate
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.normalization import BatchNormalization
def generator_model(img_dim, model_name="generator"):
x_input = Input(shape=img_dim, name="input")
x1_1 = Conv2D(64, (3, 3), strides=(2, 2), padding="same", name="1_conv")(x_input)
x1_2 = LeakyReLU(0.2, name="1_lrelu")(x1_1)
x2_1 = Conv2D(128, (3, 3), strides=(2, 2), padding="same", name="2_conv")(x1_2)
x2_2 = BatchNormalization(name="2_bn")(x2_1)
x2_3 = LeakyReLU(0.2, name="2_lrelu")(x2_2)
x3_1 = Conv2D(256, (3, 3), strides=(2, 2), padding="same", name="3_conv")(x2_3)
x3_2 = BatchNormalization(name="3_bn")(x3_1)
x3_3 = LeakyReLU(0.2, name="3_lrelu")(x3_2)
x4_1 = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="4_conv")(x3_3)
x4_2 = BatchNormalization(name="4_bn")(x4_1)
x4_3 = LeakyReLU(0.2, name="4_lrelu")(x4_2)
x5_1 = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="5_conv")(x4_3)
x5_2 = BatchNormalization(name="5_bn")(x5_1)
x5_3 = LeakyReLU(0.2, name="5_lrelu")(x5_2)
x6_1 = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="6_conv")(x5_3)
x6_2 = BatchNormalization(name="6_bn")(x6_1)
x6_3 = LeakyReLU(0.2, name="6_lrelu")(x6_2)
x7_1 = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="7_conv")(x6_3)
x7_2 = BatchNormalization(name="7_bn")(x7_1)
x7_3 = LeakyReLU(0.2, name="7_lrelu")(x7_2)
x8_1 = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="8_conv")(x7_3)
x8_2 = BatchNormalization(name="8_bn")(x8_1)
x8_3 = Activation("relu", name="8_relu")(x8_2)
x9_1 = UpSampling2D(size=(2, 2), name="9_upsample")(x8_3)
x9_2 = Conv2D(512, (3, 3), padding="same", name="9_conv")(x9_1)
x9_3 = BatchNormalization(name="9_bn")(x9_2)
x9_4 = Dropout(0.5, name="9_dropout")(x9_3)
x9_5 = Concatenate(name="9_concate")([x9_4, x7_2])
x9_6 = Activation("relu", name="9_relu")(x9_5)
x10_1 = UpSampling2D(size=(2, 2), name="10_upsample")(x9_6)
x10_2 = Conv2D(512, (3, 3), padding="same", name="10_conv")(x10_1)
x10_3 = BatchNormalization(name="10_bn")(x10_2)
x10_4 = Dropout(0.5, name="10_dropout")(x10_3)
x10_5 = Concatenate(name="10_concate")([x10_4, x6_2])
x10_6 = Activation("relu", name="10_relu")(x10_5)
x11_1 = UpSampling2D(size=(2, 2), name="11_upsample")(x10_6)
x11_2 = Conv2D(512, (3, 3), padding="same", name="11_conv")(x11_1)
x11_3 = BatchNormalization(name="11_bn")(x11_2)
x11_4 = Dropout(0.5, name="11_dropout")(x11_3)
x11_5 = Concatenate(name="11_concate")([x11_4, x5_2])
x11_6 = Activation("relu", name="11_relu")(x11_5)
x12_1 = UpSampling2D(size=(2, 2), name="12_upsample")(x11_6)
x12_2 = Conv2D(512, (3, 3), padding="same", name="12_conv")(x12_1)
x12_3 = BatchNormalization(name="12_bn")(x12_2)
x12_4 = Concatenate(name="12_concate")([x12_3, x4_2])
x12_5 = Activation("relu", name="12_relu")(x12_4)
x13_1 = UpSampling2D(size=(2, 2), name="13_upsample")(x12_5)
x13_2 = Conv2D(256, (3, 3), padding="same", name="13_conv")(x13_1)
x13_3 = BatchNormalization(name="13_bn")(x13_2)
x13_4 = Concatenate(name="13_concate")([x13_3, x3_2])
x13_5 = Activation("relu", name="13_relu")(x13_4)
x14_1 = UpSampling2D(size=(2, 2), name="14_upsample")(x13_5)
x14_2 = Conv2D(128, (3, 3), padding="same", name="14_conv")(x14_1)
x14_3 = BatchNormalization(name="14_bn")(x14_2)
x14_4 = Concatenate(name="14_concate")([x14_3, x2_2])
x14_5 = Activation("relu", name="14_relu")(x14_4)
x15_1 = UpSampling2D(size=(2, 2), name="15_upsample")(x14_5)
x15_2 = Conv2D(64, (3, 3), padding="same", name="15_conv")(x15_1)
x15_3 = BatchNormalization(name="15_bn")(x15_2)
x15_4 = Concatenate(name="15_concate")([x15_3, x1_1])
x15_5 = Activation("relu", name="15_relu")(x15_4)
x16_1 = UpSampling2D(size=(2, 2), name="16_upsample")(x15_5)
x16_2 = Conv2D(3, (3, 3), padding="same", name="16_conv")(x16_1)
x_output = Activation("tanh", name="output")(x16_2) # keep pixel value in [-1,1]
generator = Model(inputs=x_input, outputs=x_output, name=model_name)
return generator
def discriminator_model(img_dim, model_name="discriminator"):
x_input = Input(shape=img_dim, name="input")
x = Conv2D(64, (3, 3), strides=(2, 2), padding="same", name="conv_1")(x_input)
x = LeakyReLU(0.2, name="lrelu_1")(x)
x = Conv2D(128, (3, 3), strides=(2, 2), padding="same", name="conv_2")(x)
x = BatchNormalization(name="bn_2")(x)
x = LeakyReLU(0.2, name="lrelu_2")(x)
x = Conv2D(256, (3, 3), strides=(2, 2), padding="same", name="conv_3")(x)
x = BatchNormalization(name="bn_3")(x)
x = LeakyReLU(0.2, name="lrelu_3")(x)
x = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="conv_4")(x)
x = BatchNormalization(name="bn_4")(x)
x = LeakyReLU(0.2, name="lrelu_4")(x)
x = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="conv_5")(x)
x = BatchNormalization(name="bn_5")(x)
x = LeakyReLU(0.2, name="lrelu_5")(x)
x = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="conv_6")(x)
x = BatchNormalization(name="bn_6")(x)
x = LeakyReLU(0.2, name="lrelu_6")(x)
x = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="conv_7")(x)
x = BatchNormalization(name="bn_7")(x)
x = LeakyReLU(0.2, name="lrelu_7")(x)
x = Conv2D(512, (3, 3), strides=(2, 2), padding="same", name="conv_8")(x)
x = BatchNormalization(name="bn_8")(x)
x = LeakyReLU(0.2, name="lrelu_8")(x)
x_flat = Flatten(name="flatten")(x)
x_output = Dense(1, activation="sigmoid", name="output")(x_flat)
discriminator = Model(inputs=x_input, outputs=x_output, name=model_name)
return discriminator
def DCGAN(generator_model, discriminator_model, img_dim, model_name="DCGAN"):
G_input = Input(shape=img_dim, name="input")
G_output = generator_model(G_input)
D_output = discriminator_model(G_output)
DCGAN = Model(inputs=[G_input],
outputs=[G_output, D_output],
name=model_name)
DCGAN.summary()
from keras.utils import plot_model
plot_model(DCGAN, to_file="./figures/%s.png" % model_name, show_shapes=True, show_layer_names=True)
return DCGAN
def load(model_name, img_dim):
if model_name == "generator":
model = generator_model(img_dim, model_name=model_name)
print("load generator")
model.summary()
from keras.utils import plot_model
plot_model(model, to_file="./figures/%s.png" % model_name, show_shapes=True, show_layer_names=True)
return model
if model_name == "discriminator":
model = discriminator_model(img_dim, model_name=model_name)
print("load discriminator")
model.summary()
from keras.utils import plot_model
plot_model(model, to_file="./figures/%s.png" % model_name, show_shapes=True, show_layer_names=True)
return model
if __name__ == "__main__":
load("generator", (256, 256, 3))
load("discriminator", (256, 256, 3))