forked from onejiin/CycleGAN-VC2
-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodel.py
201 lines (151 loc) · 11.8 KB
/
model.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
201
import os
import tensorflow as tf
from module import * #discriminator, generator_gatedcnn
from utils import l1_loss, l2_loss, cross_entropy_loss
from datetime import datetime
class CycleGAN(object):
def __init__(self, num_features, mode = 'train',
log_dir = './log', model_name='tmp.ckpt', gen_model='generator_gatedcnn'):
self.num_features = num_features
self.input_shape = [None, num_features, None] # [batch_size, num_features, num_frames]
# Model select
# ----------------- CycleGAN-VC1 ----------------- #
if gen_model == 'CycleGAN-VC1':
self.generator = generator_gatedcnn
self.discriminator = discriminator
# ------------------ CyleGAN-VC2 ------------------ #
elif gen_model == 'CycleGAN-VC2':
self.generator = generator_gated2Dcnn
self.discriminator = discriminator_2D
elif gen_model == 'CycleGAN2_withDeconv':
self.generator = generator_gated2Dcnn_withDeconv
self.discriminator = discriminator_2D
else:
assert gen_model is None
self.mode = mode
self.build_model()
self.optimizer_initializer()
self.saver = tf.train.Saver()
self.sess = tf.Session()
self.sess.run(tf.global_variables_initializer())
if self.mode == 'train':
self.train_step = 0
self.log_dir = os.path.join(log_dir, os.path.splitext(model_name)[0])
self.writer = tf.summary.FileWriter(self.log_dir, tf.get_default_graph())
self.generator_summaries, self.discriminator_summaries = self.summary()
def build_model(self):
# ---------------- Placeholders for real training samples ---------------- #
self.input_A_real = tf.placeholder(tf.float32, shape = self.input_shape, name = 'input_A_real')
self.input_B_real = tf.placeholder(tf.float32, shape = self.input_shape, name = 'input_B_real')
# Placeholders for fake generated samples
self.input_A_fake = tf.placeholder(tf.float32, shape = self.input_shape, name = 'input_A_fake')
self.input_B_fake = tf.placeholder(tf.float32, shape = self.input_shape, name = 'input_B_fake')
# Placeholder for test samples
self.input_A_test = tf.placeholder(tf.float32, shape = self.input_shape, name = 'input_A_test')
self.input_B_test = tf.placeholder(tf.float32, shape = self.input_shape, name = 'input_B_test')
# Place holder for lambda_cycle and lambda_identity
self.lambda_cycle = tf.placeholder(tf.float32, None, name='lambda_cycle')
self.lambda_identity = tf.placeholder(tf.float32, None, name='lambda_identity')
self.lambda_feat = tf.placeholder(tf.float32, None, name='lambda_identity')
# ---------------- Placeholders for real training samples ---------------- #
# ---------------- CycleGAN, Generator ---------------- #
# Generator A->AtoB->AtoBtoA, input_A_real->generation_B->cycle_A
self.generation_B = self.generator(inputs=self.input_A_real, reuse=False, scope_name='generator_A2B')
self.cycle_A = self.generator(inputs=self.generation_B, reuse=False, scope_name='generator_B2A')
# Generator B->BtoA->BtoAtoB, input_B_real->generation_A->cycle_B
self.generation_A = self.generator(inputs=self.input_B_real, reuse=True, scope_name='generator_B2A')
self.cycle_B = self.generator(inputs=self.generation_A, reuse=True, scope_name='generator_A2B')
# ---------------- CycleGAN, Generator ---------------- #
# ---------------- Identity/Feature loss ---------------- #
self.generation_A_identity = self.generator(inputs=self.input_A_real, reuse=True, scope_name='generator_B2A')
self.generation_B_identity = self.generator(inputs=self.input_B_real, reuse=True, scope_name='generator_A2B')
# ---------------- Identity/Feature loss ---------------- #
# ---------------- CycleGAN, Discriminator ---------------- #
self.discrimination_A_fake = self.discriminator(inputs = self.generation_A, reuse = False, scope_name = 'discriminator_A')
self.discrimination_B_fake = self.discriminator(inputs = self.generation_B, reuse = False, scope_name = 'discriminator_B')
# ---------------- CycleGAN, Discriminator ---------------- #
# ---------------- Loss Define ---------------- #
# Cycle loss
self.cycle_loss = l1_loss(y = self.input_A_real, y_hat = self.cycle_A) + l1_loss(y = self.input_B_real, y_hat = self.cycle_B)
# Identity loss
self.identity_loss = l1_loss(y = self.input_A_real, y_hat = self.generation_A_identity) + l1_loss(y = self.input_B_real, y_hat = self.generation_B_identity)
# ---------------- Loss Define ---------------- #
# ---------------- Loss Calculation ---------------- #
# ================ Generator loss ================ #
# Generator wants to fool discriminator
self.generator_loss_A2B = l2_loss(y = tf.ones_like(self.discrimination_B_fake), y_hat = self.discrimination_B_fake)
self.generator_loss_B2A = l2_loss(y = tf.ones_like(self.discrimination_A_fake), y_hat = self.discrimination_A_fake)
# Merge the two generators and the cycle loss
self.generator_loss = self.generator_loss_A2B + self.generator_loss_B2A + self.lambda_cycle * self.cycle_loss + self.lambda_identity * self.identity_loss
# ================ Generator loss ================ #
# ================ Discriminator loss ================ #
self.discrimination_input_A_real = self.discriminator(inputs = self.input_A_real, reuse = True, scope_name = 'discriminator_A')
self.discrimination_input_B_real = self.discriminator(inputs = self.input_B_real, reuse = True, scope_name = 'discriminator_B')
self.discrimination_input_A_fake = self.discriminator(inputs = self.input_A_fake, reuse = True, scope_name = 'discriminator_A')
self.discrimination_input_B_fake = self.discriminator(inputs = self.input_B_fake, reuse = True, scope_name = 'discriminator_B')
# Discriminator wants to classify real and fake correctly
self.discriminator_loss_input_A_real = l2_loss(y = tf.ones_like(self.discrimination_input_A_real), y_hat = self.discrimination_input_A_real)
self.discriminator_loss_input_A_fake = l2_loss(y = tf.zeros_like(self.discrimination_input_A_fake), y_hat = self.discrimination_input_A_fake)
self.discriminator_loss_A = (self.discriminator_loss_input_A_real + self.discriminator_loss_input_A_fake) / 2
self.discriminator_loss_input_B_real = l2_loss(y = tf.ones_like(self.discrimination_input_B_real), y_hat = self.discrimination_input_B_real)
self.discriminator_loss_input_B_fake = l2_loss(y = tf.zeros_like(self.discrimination_input_B_fake), y_hat = self.discrimination_input_B_fake)
self.discriminator_loss_B = (self.discriminator_loss_input_B_real + self.discriminator_loss_input_B_fake) / 2
# Merge the two discriminators into one
self.discriminator_loss = self.discriminator_loss_A + self.discriminator_loss_B
# ================ Discriminator loss ================ #
# ---------------- Loss Calculation ---------------- #
# Categorize variables because we have to optimize the two sets of the variables separately
trainable_variables = tf.trainable_variables()
self.discriminator_vars = [var for var in trainable_variables if 'discriminator' in var.name]
self.generator_vars = [var for var in trainable_variables if 'generator' in var.name]
# ---------------- Reserved for test ---------------- #
self.generation_B_test = self.generator(inputs=self.input_A_test, reuse=True, scope_name='generator_A2B')
self.generation_A_test = self.generator(inputs=self.input_B_test, reuse=True, scope_name='generator_B2A')
# ---------------- Reserved for test ---------------- #
def optimizer_initializer(self):
self.generator_learning_rate = tf.placeholder(tf.float32, None, name = 'generator_learning_rate')
self.discriminator_learning_rate = tf.placeholder(tf.float32, None, name = 'discriminator_learning_rate')
self.discriminator_optimizer = tf.train.AdamOptimizer(learning_rate = self.discriminator_learning_rate, beta1 = 0.5).minimize(self.discriminator_loss, var_list = self.discriminator_vars)
self.generator_optimizer = tf.train.AdamOptimizer(learning_rate = self.generator_learning_rate, beta1 = 0.5).minimize(self.generator_loss, var_list = self.generator_vars)
def train(self, input_A, input_B, lambda_cycle, lambda_identity, generator_learning_rate, discriminator_learning_rate):
generation_A, generation_B, generator_loss, _, generator_summaries, generator_loss_A2B = self.sess.run(
[self.generation_A, self.generation_B, self.generator_loss, self.generator_optimizer, self.generator_summaries, self.generator_loss_A2B], \
feed_dict = {self.lambda_cycle: lambda_cycle, self.lambda_identity: lambda_identity, self.input_A_real: input_A, self.input_B_real: input_B, self.generator_learning_rate: generator_learning_rate})
self.writer.add_summary(generator_summaries, self.train_step)
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.discriminator_learning_rate: discriminator_learning_rate, self.input_A_fake: generation_A, self.input_B_fake: generation_B})
self.writer.add_summary(discriminator_summaries, self.train_step)
self.train_step += 1
return generator_loss, discriminator_loss, generator_loss_A2B
def test(self, inputs, direction):
if direction == 'A2B':
generation = self.sess.run(self.generation_B_test, feed_dict = {self.input_A_test: inputs})
elif direction == 'B2A':
generation = self.sess.run(self.generation_A_test, feed_dict = {self.input_B_test: inputs})
else:
raise Exception('Conversion direction must be specified.')
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'):
cycle_loss_summary = tf.summary.scalar('cycle_loss', self.cycle_loss)
identity_loss_summary = tf.summary.scalar('identity_loss', self.identity_loss)
generator_loss_A2B_summary = tf.summary.scalar('generator_loss_A2B', self.generator_loss_A2B)
generator_loss_B2A_summary = tf.summary.scalar('generator_loss_B2A', self.generator_loss_B2A)
generator_loss_summary = tf.summary.scalar('generator_loss', self.generator_loss)
generator_summaries = tf.summary.merge([cycle_loss_summary, identity_loss_summary, generator_loss_A2B_summary, generator_loss_B2A_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__':
model = CycleGAN(num_features = 24)
print('Graph Compile Successeded.')