/
vae.py
200 lines (143 loc) · 5.49 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
193
194
195
196
197
198
199
200
import tensorflow.keras as keras
import tensorflow as tf
from spatial_broadcast_decoder import SpatialBroadcastDecoder
from coord_conv import CoordConv2D
from config import global_config
class VAE(tf.keras.Model):
def __init__(self, latent_dim: int, name: str = 'VAE') -> None:
super(VAE, self).__init__(name=name)
self.nlayers = global_config.nlayers
self.latent_dim = latent_dim
self.layer_sizes = []
for i in range(self.nlayers):
self.layer_sizes.append(32 * 2 ** (i // 2))
self.encoder = self.encoder_network(self.latent_dim)
self.decoder = self.decoder_network(self.latent_dim)
def convolutional_layers(self, transp: bool):
layer_sizes = self.layer_sizes
if transp:
layer_sizes.reverse()
def fn(inputs):
X = inputs
for i in range(self.nlayers):
if not transp:
layer_name = f'conv-layer-{i}'
else:
layer_name = f'transp-conv-layer-{i}'
nfilters = layer_sizes[i]
X = CoordConv2D(
transp,
filters=nfilters,
kernel_size=3,
strides=2,
padding='same',
name=layer_name,
kernel_initializer='glorot_normal',
)(X)
# X = tf.debugging.check_numerics(X, 'CoordConv2D')
X = keras.layers.BatchNormalization(axis=3)(X)
# X = tf.debugging.check_numerics(X, 'BatchNorm')
X = keras.layers.LeakyReLU()(X)
# X = tf.debugging.check_numerics(X, 'LeakyReLU')
if not transp:
continue
transp_layer_name = 'conv-layer-{}'.format(self.nlayers - i - 1)
X_transp = self.encoder.get_layer(transp_layer_name)
desired_shape = X_transp.input_shape[1:-1]
cur_shape = X.shape[1:-1]
if desired_shape != cur_shape:
dx = cur_shape[0] - desired_shape[0]
dy = cur_shape[1] - desired_shape[1]
X = keras.layers.Cropping2D(cropping=((dx, 0), (dy, 0)))(X)
return X
return fn
def encoder_network(self, latent_dim: int) -> tf.keras.Model:
inputs = keras.Input(
shape=(
global_config.img_height,
global_config.img_width,
global_config.img_channels
)
)
X = inputs
X = self.convolutional_layers(False)(X)
X = keras.layers.Flatten(name='encoder-flatten')(X)
X = keras.layers.Dense(
256,
name='encoder-fc',
kernel_initializer='glorot_normal',
)(X)
# X = tf.debugging.check_numerics(X, 'FC')
X = keras.layers.LeakyReLU()(X)
# X = tf.debugging.check_numerics(X, 'FC-LeakyReLU')
mean = keras.layers.Dense(latent_dim)(X)
logvar = keras.layers.Dense(latent_dim)(X)
model = keras.models.Model(
inputs=inputs,
outputs=[mean, logvar],
name='encoder')
return model
def decoder_network(self, latent_dim: int) -> tf.keras.Model:
inputs = keras.Input(shape=(latent_dim,))
X = inputs
X = SpatialBroadcastDecoder(
global_config.img_height + 8,
global_config.img_width + 8,
)(X)
for _ in range(self.nlayers):
X = tf.keras.layers.Conv2D(
kernel_size=3,
filters=32,
padding='valid',
strides=1,
kernel_initializer='glorot_normal',
activation='relu',
)(X)
# X = tf.debugging.check_numerics(X, 'Conv2D')
img = tf.keras.layers.Conv2D(
filters=global_config.img_channels,
kernel_size=3,
activation='sigmoid',
padding='same',
name='decoder-image',
kernel_initializer='glorot_normal',
)(X)
# X = tf.debugging.check_numerics(X, 'img')
confidence = tf.keras.layers.Conv2D(
filters=1,
kernel_size=3,
padding='same',
name='decoder-raw-confidence',
kernel_initializer='glorot_normal',
)(X)
# X = tf.debugging.check_numerics(X, 'confidence')
model = keras.models.Model(
inputs=inputs, outputs=[img, confidence], name='decoder')
return model
def sample(self, eps):
return self.decode(eps)
def encode(self, x):
(mean, logvar) = self.encoder(x)
return (mean, logvar)
def reparametrize(self, mean, logvar):
eps = tf.random.normal(shape=tf.shape(mean))
return eps * tf.exp(1 / 2 * logvar) + mean
def decode(self, z):
return self.decoder(z)
@staticmethod
def compute_kl_loss(mean, logvar):
kl_loss = 1 + logvar - mean**2 - tf.exp(logvar)
kl_loss = tf.math.reduce_sum(kl_loss, axis=1)
kl_loss *= -0.5
return kl_loss
def call(self, inputs):
mean, logvar = self.encode(inputs)
self.last_mean = mean
self.last_logvar = logvar
z = self.reparametrize(mean, logvar)
return self.decode(z)
def get_trainable_variables(self):
return self.trainable_variables
def summarize(self):
self.encoder.summary()
self.decoder.summary()