/
cyclegan.py
167 lines (130 loc) · 9.1 KB
/
cyclegan.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
# CycleGan-voice convert module
import tensorflow as tf
import numpy as np
import os, datetime
import random
from Utils.networks import discriminator, generator, generator_unet
from Utils.losses import *
class CycleGAN(object) :
def __init__(self, num_features, g_type = "gated_cnn",discriminator = discriminator ,generator = generator, generator_unet = generator_unet ,mode = 'train', log_dir = './') :
self.num_features = num_features
self.input_shape = [None,num_features,None] # batch_size, num_features, num_frames
self.mode = mode
if g_type == "gated_cnn" :
self.generator = generator # gatedCNN
else :
self.generator = generator_unet
assert g_type == "u_net"
self.discriminator = discriminator
self.build_model()
self.optimizer_initializer()
self.saver = tf.train.Saver() # save checkpoint
self.sess = tf.Session()
self.sess.run(tf.global_variables_initializer())
if self.mode == 'train':
self.train_step = 0
now = datetime.datetime.now()
self.log_dir = os.path.join(log_dir, now.strftime('%Y%m%d-%H%M%S'))
self.writer = tf.summary.FileWriter(self.log_dir, tf.get_default_graph()) # Tensorboard
self.generator_summaries, self.discriminator_summaries = self.summary()
def build_model(self) :
tf.reset_default_graph()
self.input_A_real = tf.placeholder(tf.float32, shape = self.input_shape)
self.input_B_real = tf.placeholder(tf.float32, shape = self.input_shape)
self.input_A_fake = tf.placeholder(tf.float32, shape = self.input_shape)
self.input_B_fake = tf.placeholder(tf.float32, shape = self.input_shape)
self.input_A_test = tf.placeholder(tf.float32, shape = self.input_shape)
self.input_B_test = tf.placeholder(tf.float32, shape = self.input_shape)
self.generation_A = self.generator(self.input_B_real, name = "g_B2A") # G : B = > A
self.generation_B = self.generator(self.input_A_real, name = "g_A2B") # F(inverse of G) : A => B
self.cycle_A = self.generator(self.generation_B, reuse = True, name = "g_B2A") # F(g_A)
self.cycle_B = self.generator(self.generation_A, reuse = True, name = "g_A2B") # G(g_B)
# for identity loss
self.identity_A = self.generator(self.input_A_real, reuse = True, name = "g_B2A")
self.identity_B = self.generator(self.input_B_real, reuse = True, name = "g_A2B")
# generator loss
# adversarial loss
self.discrimination_A_fake = self.discriminator(self.generation_A, name = "d_A") # discriminator for A
self.discrimination_B_fake = self.discriminator(self.generation_B, name = "d_B") # discriminator for B
self.generator_loss_B2A = l2_loss(tf.ones_like(self.discrimination_A_fake), self.discrimination_A_fake)
self.generator_loss_A2B = l2_loss(tf.ones_like(self.discrimination_B_fake), self.discrimination_B_fake)
# Cycle loss
self.cycle_loss = l1_loss(self.cycle_A,self.input_A_real) + l1_loss(self.cycle_B, self.input_B_real)
self.cycle_loss_lambda = tf.placeholder(tf.float32, shape = None, name = "cycle_loss_lambda")
# Identity loss
self.identity_loss = l1_loss(self.identity_A, self.input_A_real) + l1_loss(self.identity_B, self.input_B_real)
self.identity_loss_lambda = tf.placeholder(tf.float32, None, name = "identity_loss_lambda")
# Full generator loss
self.generator_loss = (self.generator_loss_B2A +
self.generator_loss_A2B +
self.cycle_loss_lambda*self.cycle_loss +
self.identity_loss_lambda * self.identity_loss)
# Discriminator loss
self.discrimination_input_A_real = self.discriminator(self.input_A_real, reuse = True, name = "d_A")
self.discrimination_input_B_real = self.discriminator(self.input_B_real, reuse = True, name = "d_B")
self.discrimination_input_A_fake = self.discriminator(self.input_A_fake, reuse = True, name = "d_A")
self.discrimination_input_B_fake = self.discriminator(self.input_B_fake, reuse = True, name = "d_B")
self.discriminator_loss_A_real = l2_loss(tf.ones_like(self.discrimination_input_A_real),self.discrimination_input_A_real)
self.discriminator_loss_A_fake = l2_loss(tf.zeros_like(self.discrimination_input_A_fake),self.discrimination_input_A_fake)
self.discriminator_loss_B_real = l2_loss(tf.ones_like(self.discrimination_input_B_real),self.discrimination_input_B_real)
self.discriminator_loss_B_fake = l2_loss(tf.zeros_like(self.discrimination_input_B_fake),self.discrimination_input_B_fake)
self.discriminator_loss_A = self.discriminator_loss_A_real + self.discriminator_loss_A_fake
self.discriminator_loss_B = self.discriminator_loss_B_real + self.discriminator_loss_B_fake
self.discriminator_loss = self.discriminator_loss_A + self.discriminator_loss_B
trainable_variables = tf.trainable_variables()
self.discriminator_vars = [var for var in trainable_variables if 'd' in var.name]
self.generator_vars = [var for var in trainable_variables if 'g' in var.name]
def optimizer_initializer(self) :
'''linearly decay over next 200000 iter '''
self.generator_lr = tf.placeholder(tf.float32, shape = None) # default : 0.0002
self.discriminator_lr = tf.placeholder(tf.float32, shape = None) # default : 0.0001
self.generator_optimizer = tf.train.AdamOptimizer(self.generator_lr, beta1 = 0.5).minimize(self.generator_loss, var_list = self.generator_vars)
self.discriminator_optimizer = tf.train.AdamOptimizer(self.discriminator_lr, beta1 = 0.5).minimize(self.discriminator_loss, var_list = self.discriminator_vars)
def train(self,input_A,input_B,cycle_lambda,identity_lambda,generator_lr,discriminator_lr) :
# generator training
generation_A, generation_B, generator_loss, generator_summaries, _ = self.sess.run(
[self.generation_A,self.generation_B,self.generator_loss,self.generator_summaries,self.generator_optimizer],
feed_dict = {self.input_A_real : input_A, self.input_B_real : input_B,
self.cycle_loss_lambda : cycle_lambda, self.identity_loss_lambda : identity_lambda,
self.generator_lr : generator_lr})
self.writer.add_summary(generator_summaries, self.train_step)
# discriminator training
discriminator_loss, _, discriminator_summaries = self.sess.run(
[self.discriminator_loss, self.discriminator_optimizer, self.discriminator_summaries],
feed_dict = {self.input_A_real : input_A, self.input_B_real : input_B,
self.input_A_fake : generation_A, self.input_B_fake : generation_B,
self.discriminator_lr : discriminator_lr})
self.writer.add_summary(discriminator_summaries,self.train_step)
self.train_step += 1
return generator_loss, discriminator_loss
def test(self, inputs, direction) :
# Test A2B
if direction == "A2B" :
generation = self.sess.run(self.generation_B,feed_dict={self.input_A_real:inputs}) # generate B
elif direction == "B2A" :
generation = self.sess.run(self.generation_A,feed_dict={self.input_B_real:inputs}) # generate A
else :
assert (direction in ["A2B","B2A"])
return generation
def save(self, directory, filename) :
if not os.path.exists(directory):
os.makedirs(directory)
self.saver.save(self.sess, os.path.join(directory, filename))
return os.path.join(directory, filename)
def load(self, filepath):
self.saver.restore(self.sess, filepath)
def summary(self) :
with tf.name_scope("generator_summaries") :
generator_loss_B2A_summary = tf.summary.scalar("generator_loss_B2A",self.generator_loss_B2A)
generator_loss_A2B_summary = tf.summary.scalar("generator_loss_A2B",self.generator_loss_A2B)
generator_loss_summary = tf.summary.scalar("generator_loss" , self.generator_loss)
generator_summaries = tf.summary.merge([generator_loss_B2A_summary,generator_loss_A2B_summary,generator_loss_summary])
with tf.name_scope("discriminator_summaries") :
discriminator_loss_A_summary = tf.summary.scalar("discriminator_loss_A", self.discriminator_loss_A)
discriminator_loss_B_summary = tf.summary.scalar("discriminator_loss_B", self.discriminator_loss_B)
discriminator_loss_summary = tf.summary.scalar("discriminator_loss", self.discriminator_loss)
discriminator_summaries = tf.summary.merge([discriminator_loss_A_summary,discriminator_loss_B_summary,discriminator_loss_summary])
return generator_summaries, discriminator_summaries
if __name__ == "__main__" :
# the speech data were 24 mel - cepstral coefficients ( 24 MCEPS)
model = CycleGAN(num_features=24)