forked from kozistr/Awesome-GANs
-
Notifications
You must be signed in to change notification settings - Fork 0
/
acgan_model.py
187 lines (145 loc) · 6.68 KB
/
acgan_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
import tensorflow as tf
import sys
sys.path.append('../')
import tfutil as t
tf.set_random_seed(777) # reproducibility
class ACGAN:
def __init__(self, s, batch_size=100, height=32, width=32, channel=3, n_classes=10,
sample_num=10 * 10, sample_size=10,
df_dim=16, gf_dim=384, z_dim=100, lr=2e-4):
"""
# General Settings
:param s: TF Session
:param batch_size: training batch size, default 100
:param height: image height, default 32
:param width: image width, default 32
:param channel: image channel, default 3
:param n_classes: DataSet's classes, default 10
# Output Settings
:param sample_num: the number of output images, default 100
:param sample_size: sample image size, default 10
# For CNN model
:param df_dim: discriminator filter, default 16
:param gf_dim: generator filter, default 384
# Training Option
:param z_dim: z dimension (kinda noise), default 100
:param lr: learning rate, default 2e-4
"""
self.s = s
self.batch_size = batch_size
self.height = height
self.width = width
self.channel = channel
self.image_shape = [self.batch_size, self.height, self.width, self.channel]
self.n_classes = n_classes
self.sample_num = sample_num
self.sample_size = sample_size
self.df_dim = df_dim
self.gf_dim = gf_dim
self.z_dim = z_dim
self.beta1 = 0.5
self.beta2 = 0.999
self.lr = lr
# pre-defined
self.g_loss = 0.
self.d_loss = 0.
self.c_loss = 0.
self.g = None
self.g_test = None
self.d_op = None
self.g_op = None
self.c_op = None
self.merged = None
self.writer = None
self.saver = None
# Placeholders
self.x = tf.placeholder(tf.float32,
shape=[None, self.height, self.width, self.channel],
name="x-image") # (-1, 32, 32, 3)
self.y = tf.placeholder(tf.float32, shape=[None, self.n_classes], name="y-label") # (-1, 10)
self.y_rnd = tf.placeholder(tf.float32, shape=[None, self.n_classes], name="y-rnd-label") # (-1, 10)
self.z = tf.placeholder(tf.float32, shape=[None, self.z_dim], name="z-noise") # (-1, 100)
self.build_acgan() # build ACGAN model
def discriminator(self, x, reuse=None):
"""
# Following a D Network, CiFar-like-hood, referred in the paper
:param x: images
:param y: labels
:param reuse: re-usable
:return: classification, probability (fake or real), network
"""
with tf.variable_scope("discriminator", reuse=reuse):
x = t.conv2d(x, self.df_dim, 3, 2, name='disc-conv2d-1')
x = tf.nn.leaky_relu(x, alpha=0.2)
x = tf.layers.dropout(x, 0.5, name='disc-dropout2d-1')
for i in range(5):
x = t.conv2d(x, self.df_dim * (2 ** (i + 1)), k=3, s=(i % 2 + 1), name='disc-conv2d-%d' % (i + 2))
x = t.batch_norm(x, reuse=reuse, name="disc-bn-%d" % (i + 1))
x = tf.nn.leaky_relu(x, alpha=0.2)
x = tf.layers.dropout(x, 0.5, name='disc-dropout2d-%d' % (i + 1))
net = tf.layers.flatten(x)
cat = t.dense(net, self.n_classes, name='disc-fc-cat')
disc = t.dense(net, 1, name='disc-fc-disc')
return cat, disc, net
def generator(self, z, y, reuse=None, is_train=True):
"""
# Following a G Network, CiFar-like-hood, referred in the paper
:param z: noise
:param y: image label
:param reuse: re-usable
:param is_train: trainable
:return: prob
"""
with tf.variable_scope("generator", reuse=reuse):
x = tf.concat([z, y], axis=1) # (-1, 110)
x = t.dense(x, self.gf_dim, name='gen-fc-1')
x = tf.nn.relu(x)
x = tf.reshape(x, (-1, 4, 4, 24))
for i in range(1, 3):
x = t.deconv2d(x, self.gf_dim // (2 ** i), 5, 2, name='gen-deconv2d-%d' % (i + 1))
x = t.batch_norm(x, is_train=is_train, reuse=reuse, name="gen-bn-%d" % i)
x = tf.nn.relu(x)
x = t.deconv2d(x, self.channel, 5, 2, name='gen-deconv2d-4')
x = tf.nn.tanh(x) # scaling to [-1, 1]
return x
def build_acgan(self):
# Generator
self.g = self.generator(self.z, self.y)
# Discriminator
c_real, d_real, _ = self.discriminator(self.x)
c_fake, d_fake, _ = self.discriminator(self.g, reuse=True)
# sigmoid ce loss
d_real_loss = t.sce_loss(d_real, tf.ones_like(d_real))
d_fake_loss = t.sce_loss(d_fake, tf.zeros_like(d_fake))
self.d_loss = d_real_loss + d_fake_loss
self.g_loss = t.sce_loss(d_fake, tf.ones_like(d_fake))
# softmax ce loss
c_real_loss = t.softce_loss(c_real, self.y)
c_fake_loss = t.softce_loss(c_fake, self.y)
self.c_loss = c_real_loss + c_fake_loss
# self.d_loss = self.d_loss + self.c_loss
# self.g_loss = self.g_loss - self.c_loss
# Summary
tf.summary.scalar("loss/d_real_loss", d_real_loss)
tf.summary.scalar("loss/d_fake_loss", d_fake_loss)
tf.summary.scalar("loss/d_loss", self.d_loss)
tf.summary.scalar("loss/c_real_loss", c_real_loss)
tf.summary.scalar("loss/c_fake_loss", c_fake_loss)
tf.summary.scalar("loss/c_loss", self.c_loss)
tf.summary.scalar("loss/g_loss", self.g_loss)
# Optimizer
t_vars = tf.trainable_variables()
d_params = [v for v in t_vars if v.name.startswith('d')]
g_params = [v for v in t_vars if v.name.startswith('g')]
c_params = [v for v in t_vars if v.name.startswith('d') or v.name.startswith('g')]
self.d_op = tf.train.AdamOptimizer(self.lr,
beta1=self.beta1, beta2=self.beta2).minimize(self.d_loss, var_list=d_params)
self.g_op = tf.train.AdamOptimizer(self.lr,
beta1=self.beta1, beta2=self.beta2).minimize(self.g_loss, var_list=g_params)
self.c_op = tf.train.AdamOptimizer(self.lr,
beta1=self.beta1, beta2=self.beta2).minimize(self.c_loss, var_list=c_params)
# Merge summary
self.merged = tf.summary.merge_all()
# Model saver
self.saver = tf.train.Saver(max_to_keep=1)
self.writer = tf.summary.FileWriter('./model/', self.s.graph)