-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathGAN.py
187 lines (137 loc) · 7.08 KB
/
GAN.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
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2'
import tensorflow as tf
import numpy as np
import tensorflow.contrib.slim as slim
from tensorflow.examples.tutorials.mnist import input_data
from util import *
batch_size = 64
visualize_dim = 196 #number of images to be sampled and stored in the outputDir
display_step = 100
modelsDir = './model1'
outputDir = './images1'
model_save_step = 5000
save_imageSamples_step = 500 #number of steps after images have to be saved
n_epochs = 50000
inputDataDir = 'MNIST_data/'
if not os.path.exists(modelsDir) :
os.mkdir(modelsDir)
if not os.path.exists(outputDir) :
os.mkdir(outputDir)
def lrelu(x, leak=0.2, name="lrelu"):
return tf.maximum(x, leak*x)
def generator(z,y,train = True,scope = "generator") :
with tf.variable_scope(scope) as sc :
with slim.arg_scope([slim.conv2d_transpose,slim.fully_connected],
biases_initializer=tf.constant_initializer(0.0),
activation_fn=None):
with slim.arg_scope([slim.batch_norm],decay = 0.95,center = True,scale = True,activation_fn = tf.nn.relu,is_training = train) :
yb = tf.reshape(y,[-1,1,1,10])
z = tf.concat([z,y],1)
net = slim.fully_connected(z,1024,scope = "linear_1")
net = slim.batch_norm(net,scope = "bn_1")
net = tf.concat([net,y],1)
net = slim.batch_norm(slim.fully_connected(net,128*7*7,scope= "linear_2"),scope = "bn_2")
net = tf.reshape(net,[-1,7,7,128])
net = tf.concat([net,yb*tf.ones([batch_size,7,7,10])],3)
net = slim.conv2d_transpose(net,64,[5,5],stride = [2,2],padding = 'SAME',scope = 'conv_transpose1')
net = slim.batch_norm(net,scope = 'bn_3')
net = tf.concat([net,yb*tf.ones([batch_size,14,14,10])],3)
net = tf.nn.sigmoid(slim.conv2d_transpose(net,1,[5,5],stride = [2,2],scope = 'conv_transpose_2'))
return net
def sampler(z,y,train = True,scope = "generator") :
with tf.variable_scope(scope,reuse = True) as sc :
with slim.arg_scope([slim.conv2d_transpose,slim.fully_connected],
biases_initializer=tf.constant_initializer(0.0),
activation_fn=None):
with slim.arg_scope([slim.batch_norm],decay = 0.95,center = True,scale = True,activation_fn = tf.nn.relu,is_training = train) :
yb = tf.reshape(y,[-1,1,1,10])
z = tf.concat([z,y],1)
net = slim.fully_connected(z,1024,scope = "linear_1")
net = slim.batch_norm(net,scope = "bn_1")
net = tf.concat([net,y],1)
net = slim.batch_norm(slim.fully_connected(net,128*7*7,scope= "linear_2"),scope = "bn_2")
net = tf.reshape(net,[-1,7,7,128])
net = tf.concat([net,yb*tf.ones([visualize_dim,7,7,10])],3)
net = slim.conv2d_transpose(net,64,[5,5],stride = [2,2],padding = 'SAME',scope = 'conv_transpose1')
net = slim.batch_norm(net,scope = 'bn_3')
net = tf.concat([net,yb*tf.ones([visualize_dim,14,14,10])],3)
net = tf.nn.sigmoid(slim.conv2d_transpose(net,1,[5,5],stride = [2,2],scope = 'conv_transpose_2'))
return net
def discriminator(image,y,train = True,scope = "discriminator",reuse = False) :
with tf.variable_scope(scope,reuse = reuse) as sc :
with slim.arg_scope([slim.conv2d,slim.fully_connected],
biases_initializer=tf.constant_initializer(0.0),
activation_fn=None):
with slim.arg_scope([slim.batch_norm],decay = 0.95,center = True,scale = True,activation_fn = None,is_training = train) :
yb = tf.reshape(y,[-1,1,1,10])
x = tf.concat([image,yb*tf.ones([batch_size,28,28,10])],3)
net = lrelu(slim.conv2d(x,64,[5,5],stride = [2,2],scope = 'conv_1'))
net = tf.concat([net,yb*tf.ones([batch_size,14,14,10])],3)
net = lrelu(slim.batch_norm(slim.conv2d(net,128,[5,5],stride = [2,2],scope = 'conv_2'),scope = 'bn_1'))
net = slim.flatten(net)
net = tf.concat([net,y],1)
net = slim.fully_connected(net,1024,scope = 'linear_1')
net = lrelu(slim.batch_norm(net,scope = 'bn_2'))
net = slim.fully_connected(net,1,scope = 'linear_2')
return net
def build_model(x,z,y) :
image_gen = generator(z,y)
real_out = discriminator(x,y)
fake_out = discriminator(image_gen,y,reuse = True)
t_vars = tf.trainable_variables()
gen_vars = [var for var in t_vars if 'generator' in var.name]
disc_vars = [var for var in t_vars if 'discriminator' in var.name]
return real_out,fake_out,gen_vars,disc_vars
def getLoss(logit,label) :
return tf.nn.sigmoid_cross_entropy_with_logits(logits=logit, labels=label)
def getLosses(real_logits,fake_logits) :
d_real_loss = tf.reduce_mean(getLoss(real_logits,tf.ones_like(real_logits)))
d_fake_loss = tf.reduce_mean(getLoss(fake_logits,tf.zeros_like(fake_logits)))
g_loss = tf.reduce_mean(getLoss(fake_logits,tf.ones_like(fake_logits)))
d_loss = d_real_loss + d_fake_loss
return d_loss,g_loss
y = tf.placeholder(tf.float32, shape=[None, 10])
z = tf.placeholder(tf.float32, shape=[None, 100])
image_gen = generator(z,y)
fake_logits = discriminator(image_gen,y)
g_loss = tf.reduce_mean(getLoss(fake_logits,tf.ones_like(fake_logits)))
x = tf.placeholder(tf.float32, shape=[None, 28,28,1])
real_logits = discriminator(x,y,reuse = True)
# real_logits,fake_logits,gen_vars,disc_vars = build_model(x,z,y)
t_vars = tf.trainable_variables()
gen_vars = [var for var in t_vars if 'generator' in var.name]
disc_vars = [var for var in t_vars if 'discriminator' in var.name]
g_optim = tf.train.AdamOptimizer(2e-4, beta1=0.5).minimize(g_loss, var_list=gen_vars)
g_summary = tf.summary.scalar('generator_loss', g_loss)
d_real_loss = tf.reduce_mean(getLoss(real_logits,tf.ones_like(real_logits)))
d_fake_loss = tf.reduce_mean(getLoss(fake_logits,tf.zeros_like(fake_logits)))
d_loss = d_real_loss + d_fake_loss
d_summary = tf.summary.scalar('discriminator_loss', d_loss)
d_optim = tf.train.AdamOptimizer(2e-4, beta1=0.5).minimize(d_loss, var_list=disc_vars)
sampled = sampler(z,y)
sess = tf.InteractiveSession()
train_writer = tf.summary.FileWriter('/tmp/gan_logs', sess.graph)
sess.run(tf.global_variables_initializer())
mnist = input_data.read_data_sets(inputDataDir, one_hot=True)
sample_z = np.random.uniform(-1, 1, size=(visualize_dim, 100)).astype(np.float32)
sample_x,sample_y = mnist.train.next_batch(visualize_dim)
sample_x = np.reshape(sample_x,[-1,28,28,1])
saver = tf.train.Saver()
for epoch in range(n_epochs) :
data_x,data_y = mnist.train.next_batch(batch_size)
data_x = np.reshape(data_x,[-1,28,28,1])
z_noise = np.random.uniform(-1, 1, size=(batch_size, 100)).astype(np.float32)
_,summary = sess.run([d_optim,d_summary],feed_dict = {x:data_x,y:data_y,z:z_noise})
train_writer.add_summary(summary, epoch)
_,summary = sess.run([g_optim,g_summary],feed_dict = {y:data_y,z:z_noise})
train_writer.add_summary(summary, epoch)
_,summary = sess.run([g_optim,g_summary],feed_dict = {y:data_y,z:z_noise})
train_writer.add_summary(summary, epoch)
if (np.mod(epoch,display_step) == 1) :
print('%d steps reached'%epoch)
if (np.mod(epoch,save_imageSamples_step) == 1) :
samples = sess.run(sampled,feed_dict = {y:sample_y,z:sample_z})
save_images(samples, image_manifold_size(samples.shape[0]), os.path.join(outputDir,'image_'+str(epoch)+'.png'))
if(np.mod(epoch,model_save_step) == 1) and not epoch == 1:
saver.save(sess,os.path.join(modelsDir,'gan_model_'+str(epoch)+'.ckpt'))