-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathVAE.py
192 lines (161 loc) · 8.33 KB
/
VAE.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
import pydot
import numpy as np
import matplotlib.pyplot as plt
from keras.utils.vis_utils import plot_model
from scipy.stats import norm
import keras
from keras.optimizers import Adam
import keras.backend as bk
from keras.layers import Input, Flatten, Reshape, Dense, Lambda, Conv2DTranspose, Conv2D
from keras.layers import BatchNormalization, Activation, ZeroPadding2D
from keras.layers.advanced_activations import LeakyReLU
from keras.models import Model, Sequential, load_model
from keras import backend as K
from keras import metrics
from keras.datasets import mnist
from keras.utils import to_categorical
import os
img_rows = 28
img_cols = 28
channels = 1
img_shape = (img_rows, img_cols, channels)
latent_dim = 2
class VAE():
def __init__(self):
self.img_rows = img_rows
self.img_cols = img_cols
self.channels = channels
# 28,28,1
self.img_shape = img_shape
self.latent_dim = latent_dim
optimizer = Adam(learning_rate=0.0005)
self.encoder = self.build_encoder()
self.decoder = self.build_decoder()
VAE_inp = Input(shape=self.img_shape, name='VAE_input')
encoder_output = self.encoder(VAE_inp)
decoder_output = self.decoder(encoder_output)
self.vae_model = Model(VAE_inp, decoder_output, name='VAE')
self.vae_model.summary()
self.vae_model.compile(optimizer=optimizer, loss=self.loss_func()) # self.loss_func()
def sampling(self, mu_log_variance):
mu, log_variance = mu_log_variance
epsilon = bk.random_normal(shape=bk.shape(mu), mean=0.0, stddev=1.0)
random_sample = mu + bk.exp(log_variance / 2) * epsilon
return random_sample
def loss_func(self):
def vae_reconstruction_loss(y_true, y_predict):
reconstruction_loss_factor = 1000
reconstruction_loss = bk.mean(bk.square(y_true - y_predict), axis=[1, 2, 3])
return reconstruction_loss_factor * reconstruction_loss
def vae_kl_loss(y_true, y_predict):
kl_loss = -0.5 * bk.sum(1.0 + y_predict - bk.square(y_true) - bk.exp(y_predict), axis=1)
return kl_loss
# def vae_kl_loss_metric(y_true, y_predict):
# kl_loss = -0.5 * bk.sum(1.0 + self.log - bk.square(self.mu) - bk.exp(self.log), axis=1)
# return kl_loss
def vae_loss(y_true, y_predict):
reconstruction_loss = vae_reconstruction_loss(y_true, y_predict)
kl_loss = vae_kl_loss(y_true, y_predict)
loss = reconstruction_loss + kl_loss
return loss
return vae_loss
def build_encoder(self):
inp = Input(shape=self.img_shape, name="encoder_input")
conv1 = Conv2D(filters=1, kernel_size=(3, 3), padding="same", strides=1, name="encoder_conv_1")(inp)
norm1 = BatchNormalization(name="encoder_norm_1")(conv1)
leakyrelu1 = LeakyReLU(name="encoder_leakyrelu_1")(norm1)
conv2 = Conv2D(filters=32, kernel_size=(3, 3), padding="same", strides=1, name="encoder_conv_2")(leakyrelu1)
norm2 = BatchNormalization(name="encoder_norm_2")(conv2)
leakyrelu2 = LeakyReLU(name="encoder_leakyrelu_2")(norm2)
conv3 = Conv2D(filters=64, kernel_size=(3, 3), padding="same", strides=2, name="encoder_conv_3")(leakyrelu2)
norm3 = BatchNormalization(name="encoder_norm_3")(conv3)
leakyrelu3 = LeakyReLU(name="encoder_leakyrelu_3")(norm3)
conv4 = Conv2D(filters=64, kernel_size=(3, 3), padding="same", strides=2, name="encoder_conv_4")(leakyrelu3)
norm4 = BatchNormalization(name="encoder_norm_4")(conv4)
leakyrelu4 = LeakyReLU(name="encoder_leakyrelu_4")(norm4)
conv5 = Conv2D(filters=64, kernel_size=(3, 3), padding="same", strides=1, name="encoder_conv_5")(leakyrelu4)
norm5 = BatchNormalization(name="encoder_norm_5")(conv5)
leakyrelu5 = LeakyReLU(name="encoder_leakyrelu_5")(norm5)
self.shape_before_flatten = bk.int_shape(leakyrelu5)[1:]
flatten = Flatten(name='encoder_flatten')(leakyrelu5)
self.mu = Dense(units=self.latent_dim, name="encoder_mu")(flatten)
self.log = Dense(units=self.latent_dim, name="encoder_log_variance")(flatten)
# model = Model(inp, (mu, log), name='VAE_encoder')
output = Lambda(self.sampling, name='encoder_output')([self.mu, self.log])
model = Model(inp, output, name='VAE_encoder')
plot_model(model, show_shapes=True, to_file='./images/VAE/VAE_encoder.png')
model.summary()
return model
def build_decoder(self):
inp = Input(shape=self.latent_dim, name="decoder_input")
dense1 = Dense(units=np.prod(self.shape_before_flatten), name="decoder_dense_1")(inp)
resh = Reshape(self.shape_before_flatten)(dense1)
conv_trans1 = Conv2DTranspose(filters=64, kernel_size=(3, 3), padding="same", strides=1,
name="decoder_conv_tran_1")(resh)
norm_layer1 = BatchNormalization(name="decoder_norm_1")(conv_trans1)
leakyrelu1 = LeakyReLU(name="decoder_leakyrelu_1")(norm_layer1)
conv_trans2 = Conv2DTranspose(filters=64, kernel_size=(3, 3), padding="same", strides=2,
name="decoder_conv_tran_2")(leakyrelu1)
norm_layer2 = BatchNormalization(name="decoder_norm_2")(conv_trans2)
leakyrelu2 = LeakyReLU(name="decoder_leakyrelu_2")(norm_layer2)
conv_trans3 = Conv2DTranspose(filters=64, kernel_size=(3, 3), padding="same", strides=2,
name="decoder_conv_tran_3")(leakyrelu2)
norm_layer3 = BatchNormalization(name="decoder_norm_3")(conv_trans3)
leakyrelu3 = LeakyReLU(name="decoder_leakyrelu_3")(norm_layer3)
conv_trans4 = Conv2DTranspose(filters=1, kernel_size=(3, 3), padding="same", strides=1,
name="decoder_conv_tran_4")(leakyrelu3)
norm_layer4 = BatchNormalization(name="decoder_norm_4")(conv_trans4)
output = LeakyReLU(name="decoder_leakyrelu_4")(norm_layer4)
model = Model(inp, output, name='VAE_decoder')
plot_model(model, show_shapes=True, to_file='./images/VAE/VAE_decoder.png')
model.summary()
return model
def train(self, epochs, x_train, batch_size=64):
his = self.vae_model.fit(x_train, x_train, epochs=epochs, batch_size=batch_size, shuffle=True,
validation_split=0.2)
self.encoder.save("VAE_encoder.h5")
self.decoder.save("VAE_decoder.h5")
self.vae_model.save("VAE.h5")
return his
def predict(self, x_test):
pred = self.vae_model.predict(x_test)
self.sample_images(pred)
return pred
def sample_images(self, pred):
r, c = 5, 5
fig, axs = plt.subplots(r, c)
cnt = 0
for i in range(r):
for j in range(c):
axs[i, j].imshow(pred[cnt, :, :, 0], cmap='gray')
axs[i, j].axis('off')
cnt += 1
fig.savefig("images/VAE/vae.png")
plt.close()
if __name__ == '__main__':
if not os.path.exists("./images/VAE"):
os.makedirs("./images/VAE")
(X_train, Y_train), (X_test, Y_test) = mnist.load_data()
X_train = X_train.astype('float32') / 255.
X_test = X_test.astype('float32') / 255.
X_train = X_train.reshape((X_train.shape[0], X_train.shape[1], X_train.shape[2], 1))
X_test = X_test.reshape((X_test.shape[0], X_train.shape[1], X_train.shape[2], 1))
X_total = np.concatenate((X_train, X_test), axis=0)
# if not os.path.exists("VAE.h5"):
vae = VAE()
his = vae.train(x_train=X_total, epochs=50, batch_size=200)
vae.predict(X_train[:25])
print('loss: ' + str(np.mean(his.history['loss'])) + ', val_loss:' + str(np.mean(his.history['val_loss'])))
plt.figure()
plt.plot(his.epoch, his.history['loss'], label='loss')
plt.plot(his.epoch, his.history['val_loss'], label='val_loss')
plt.xlabel("epochs")
plt.ylabel("loss")
plt.legend()
plt.savefig("images/VAE/vae_loss.png")
'''
else:
encoder = load_model("VAE_encoder.h5", compile=False)
decoder = load_model("VAE_decoder.h5", compile=False)
vae = load_model("VAE.h5", compile=False)
'''