forked from aangfanboy/aldrin_vae
-
Notifications
You must be signed in to change notification settings - Fork 0
/
VAEmodel.py
163 lines (128 loc) · 7.67 KB
/
VAEmodel.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
import tensorflow as tf
class Barney(tf.keras.Model):
def compute_output_signature(self, input_signature):
pass
def encoder_model(self, activation_function: tf.nn = tf.nn.elu, last_layer_activation: tf.nn = None):
input_layer = tf.keras.layers.Input(shape=(self.xs, self.ys, self.channels), name="input_1")
x = tf.keras.layers.Conv2D(32, (3, 3), strides=1, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(input_layer)
x = tf.keras.layers.Dropout(0.25)(x)
x = tf.keras.layers.Conv2D(64, (3, 3), strides=2, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Conv2D(64, (3, 3), strides=1, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.MaxPooling2D((3, 3), strides=2)(x)
x = tf.keras.layers.Conv2D(128, (3, 3), strides=1, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Dropout(0.25)(x)
x = tf.keras.layers.Conv2D(256, (3, 3), strides=2, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Conv2D(256, (3, 3), strides=1, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.MaxPooling2D((3, 3), strides=2)(x)
x = tf.keras.layers.Conv2D(256, (3, 3), strides=1, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Dropout(0.25)(x)
x = tf.keras.layers.Conv2D(512, (3, 3), strides=1, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Flatten()(x)
x = tf.keras.layers.Dense(self.last_layer_units*2, activation=last_layer_activation,
kernel_regularizer=tf.keras.regularizers.l2())(x)
model = tf.keras.models.Model(input_layer, x)
model.summary()
return model
def decoder_model(self, activation_function: tf.nn = tf.nn.elu, last_layer_activation: tf.nn = None):
input_layer = tf.keras.layers.Input(
shape=(self.last_layer_units, ), name="input_2"
)
x = tf.keras.layers.Dense(512, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(input_layer)
x = tf.keras.layers.Reshape((1, 1, 512))(x)
x = tf.keras.layers.Conv2DTranspose(512, (3, 3), strides=2, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.UpSampling2D()(x)
x = tf.keras.layers.Conv2DTranspose(256, (3, 3), strides=1, activation=activation_function,
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Conv2DTranspose(256, (3, 3), strides=2, activation=activation_function, padding="same",
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Dropout(0.25)(x)
x = tf.keras.layers.Conv2DTranspose(128, (3, 3), strides=2, activation=activation_function, padding="same",
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.UpSampling2D()(x)
x = tf.keras.layers.Conv2DTranspose(64, (3, 3), strides=1, activation=activation_function, padding="same",
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Conv2DTranspose(64, (3, 3), strides=1, activation=activation_function, padding="same",
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Dropout(0.25)(x)
x = tf.keras.layers.Conv2DTranspose(32, (3, 3), strides=2, activation=activation_function, padding="same",
kernel_regularizer=tf.keras.regularizers.l2())(x)
x = tf.keras.layers.Conv2DTranspose(self.channels, (3, 3), strides=1, activation=last_layer_activation,
padding="same")(x)
model = tf.keras.models.Model(input_layer, x)
model.summary()
return model
def save_models(self):
self.encoder.save(self.file_path.replace(".h5", "_encoder.h5"))
self.decoder.save(self.file_path.replace(".h5", "_decoder.h5"))
def __init__(self, image_shape: tuple, file_path: str, last_layer_units: int = 1024, lr: float = 0.001):
super(Barney, self).__init__()
self.xs, self.ys, self.channels = image_shape
self.file_path = file_path
self.last_layer_units, self.lr = last_layer_units, lr
tf.compat.v1.gfile.MakeDirs("".join(self.file_path.split("/")[:-1]))
try:
self.encoder = tf.keras.models.load_model(self.file_path.replace(".h5", "_encoder.h5"),
custom_objects={"leaky_relu": tf.nn.leaky_relu})
except OSError:
self.encoder = self.encoder_model(
activation_function=tf.nn.leaky_relu,
last_layer_activation=None
)
try:
self.decoder = tf.keras.models.load_model(self.file_path.replace(".h5", "_decoder.h5"),
custom_objects={"leaky_relu": tf.nn.leaky_relu})
except OSError:
self.decoder = self.decoder_model(
activation_function=tf.nn.leaky_relu,
last_layer_activation=None
)
self.optimizer = tf.keras.optimizers.Adam(self.lr, beta_1=0.5)
def encode(self, x: tf.Tensor):
mean, logvar = tf.split(self.encoder(x), num_or_size_splits=2, axis=1)
return mean, logvar
def decode(self, z: tf.Tensor, apply_sigmoid: bool = False):
logits = self.decoder(z)
if apply_sigmoid:
logits = tf.sigmoid(logits)
return logits
@staticmethod
def reparameterize(mean: tf.Tensor, logvar: float):
return tf.random.normal(shape=mean.shape) * tf.exp(logvar * .5) + mean
def generate_sample(self, eps: tf.Tensor):
return self.decode(eps, apply_sigmoid=True)
@staticmethod
def log_normal_pdf(sample: float, mean: float, logvar: float, raxis: float = 1):
return tf.reduce_sum(-.5 * ((sample - mean) ** 2. * tf.exp(-logvar) + logvar + 1.837877), axis=raxis)
@tf.function
def compute_loss(self, x: tf.Tensor):
mean, logvar = self.encode(x)
z = self.reparameterize(mean, logvar)
x_logit = self.decode(z)
cross_ent = tf.nn.sigmoid_cross_entropy_with_logits(logits=x_logit, labels=x)
logpx_z = -tf.reduce_sum(cross_ent, axis=[1, 2, 3])
logpz = self.log_normal_pdf(z, 0., 0.)
logqz_x = self.log_normal_pdf(z, mean, logvar)
return -tf.reduce_mean(logpx_z + logpz - logqz_x), x_logit
@tf.function
def train_step(self, x: tf.Tensor):
with tf.GradientTape() as tape:
loss, outputs = self.compute_loss(x)
gradients = tape.gradient(loss, self.trainable_variables)
self.optimizer.apply_gradients(zip(gradients, self.trainable_variables))
return loss, outputs
if __name__ == '__main__':
barney = Barney(
image_shape=(128, 128, 3),
file_path="models/my_model.h5",
lr=0.0001
)