/
version3.py
286 lines (274 loc) · 13.4 KB
/
version3.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
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
# -*- coding: utf-8 -*-
"""
Created on Tue Oct 23 17:02:16 2018
@author: wmy
"""
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
import os
import random
import time
import math
class DCGAN(object):
'''
Arguments:
dataset_floder: the path of dataset
image_height: output images' height
image_width: output images' width
batch_size: mini batch size
max_trainset_size: max size of trainset
z_dim: the len of z
generator_filters_list: number of filters in generator's layers
generator_kernel_size_list: kernel size of filters in generator's layers
generator_kernel_strides_list: strides in conv2d
discriminator_filters_list: number of filters in discriminator's layers
discriminator_kernel_size_list: kernel size of filters in discriminator's layers
discriminator_strides_list: strides in conv2d
learning_rate: learning rate
leaky_relu_alpha: leaky relu alpha
adam_beta1: adam beta1
epoch: num of epoches
name: A string, name of the DCGAN
images_save_floder: images save path
n_plot_height: number of images in the height of plot images
n_plot_width: number of images in the width of plot images
plot_images_name: plot images name
'''
def __init__(self,
dataset_floder,
image_height=64,
image_width=64,
batch_size=64,
max_trainset_size=1024,
z_dim=128,
generator_filters_list=[1024, 512, 256, 128],
generator_kernel_size_list=[[5,5], [5,5], [5,5], [5,5]],
generator_kernel_strides_list=[(2,2), (2,2), (2,2), (2,2)],
discriminator_filters_list=[64, 128, 256, 512],
discriminator_kernel_size_list=[[5,5], [5,5], [5,5], [5,5]],
discriminator_strides_list=[(2,2), (2,2), (2,2), (2,2)],
learning_rate=0.0001,
leaky_relu_alpha=0.2,
adam_beta1=0.5,
epoch=500,
name=None,
images_save_floder='./results',
n_plot_height=None,
n_plot_width=None,
plot_images_name="DCGAN_GEN_"):
assert(len(generator_filters_list)==len(generator_kernel_size_list) and \
len(generator_filters_list)==len(generator_kernel_strides_list))
assert(len(discriminator_filters_list)==len(discriminator_kernel_size_list) and \
len(discriminator_filters_list)==len(discriminator_strides_list))
self.dataset_floder = dataset_floder
self.image_height = image_height
self.image_width = image_width
self.batch_size = batch_size
self.max_trainset_size = max_trainset_size
self.z_dim = z_dim
self.generator_filters_list = generator_filters_list
self.generator_kernel_size_list = generator_kernel_size_list
self.generator_kernel_strides_list = generator_kernel_strides_list
self.discriminator_filters_list = discriminator_filters_list
self.discriminator_kernel_size_list = discriminator_kernel_size_list
self.discriminator_strides_list = discriminator_strides_list
self.learning_rate = learning_rate
self.leaky_relu_alpha = leaky_relu_alpha
self.adam_beta1 = adam_beta1
self.epoch = epoch
self.name = name
self.images_save_floder = images_save_floder
self.n_plot_height = n_plot_height
self.n_plot_width = n_plot_width
self.plot_images_name = plot_images_name
self.make_dir(images_save_floder)
pass
def load_trainset(self):
'''load trainset from the floder'''
floder = self.dataset_floder
# names of your train images
images = os.listdir(floder)
# the number of train images
num_images = min(len(images), self.max_trainset_size)
resize_height = self.image_height
resize_width = self.image_width
dataset = np.empty((num_images, resize_height, resize_width, 3), dtype="float32")
for i in range(num_images):
img = Image.open(floder + "/" + images[i])
# size (w, h)
# shape (h, w)
img = img.resize((resize_width, resize_height))
img_arr = np.asarray(img, dtype="float32")
dataset[i, :, :, :] = img_arr
pass
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
dataset = tf.reshape(dataset, [-1, resize_height, resize_width, 3])
# range to -1, 1
traindata = dataset * 1.0 / 127.5 - 1.0
# one vector
traindata = tf.reshape(traindata, [-1, resize_height*resize_width*3])
trainset = sess.run(traindata)
print('[OK] ' + str(num_images) + ' samples have been loaded')
return trainset
def generator(self, z, reuse, trainable=True):
'''creat a generator'''
h_init = self.image_height
w_init = self.image_width
for i in range(len(self.generator_kernel_strides_list)):
(h_stride, w_stride) = self.generator_kernel_strides_list[i]
h_init = h_init / h_stride
w_init = w_init / w_stride
pass
h_init = int(h_init)
w_init = int(w_init)
with tf.variable_scope("generator", reuse=reuse):
# layer 1: FC
output = tf.layers.dense(z, self.generator_filters_list[0]*h_init*w_init, \
trainable=trainable)
output = tf.reshape(output, [self.batch_size, h_init, w_init, \
self.generator_filters_list[0]])
output = tf.layers.batch_normalization(output, training=trainable)
output = tf.nn.relu(output)
# layer 2 to L-1
for i in range(1, len(self.generator_filters_list)):
filters = self.generator_filters_list[i]
kernel_size = self.generator_kernel_size_list[i-1]
strides = self.generator_kernel_strides_list[i-1]
output = tf.layers.conv2d_transpose(output, filters, kernel_size, strides=strides, \
padding="SAME", trainable=trainable)
output = tf.layers.batch_normalization(output, training=trainable)
output = tf.nn.relu(output)
pass
# layer L
kernel_size = self.generator_kernel_size_list[-1]
strides = self.generator_kernel_strides_list[-1]
output = tf.layers.conv2d_transpose(output, 3, kernel_size, strides=strides, \
padding="SAME", trainable=trainable)
generator_images = tf.nn.tanh(output)
return generator_images
def discriminator(self, x, reuse, trainable=True):
'''creat a discriminator'''
output = x
with tf.variable_scope("discriminator", reuse=reuse):
for i in range(len(self.discriminator_filters_list)-1):
filters = self.discriminator_filters_list[i]
kernel_size = self.discriminator_kernel_size_list[i]
strides = self.discriminator_strides_list[i]
output = tf.layers.conv2d(x, filters, kernel_size, strides=strides, \
padding="SAME", trainable=trainable)
output = tf.layers.batch_normalization(output, training=trainable)
output = tf.nn.leaky_relu(output, alpha=self.leaky_relu_alpha)
pass
output = tf.layers.flatten(output)
discriminator_output = tf.layers.dense(output, 1, trainable=trainable)
return discriminator_output
def make_dir(self, path):
'''make the path'''
folder = os.path.exists(path)
if not folder:
os.makedirs(path)
print("You created a new path!")
print("Path: " + str(path))
pass
else:
print("Path: " + str(path) + " is already existed.")
pass
def plot_save_outputs(self, index, images):
'''plot and save images'''
n_height = self.n_plot_height
n_width = self.n_plot_width
batch_size = self.batch_size
h = 0
w = 0
if n_height==None and n_width==None:
h = np.int(np.sqrt(batch_size))
w = np.int(np.sqrt(batch_size))
elif n_height!=None and n_width==None:
h = n_height
w = np.int(1.0*batch_size/h)
elif n_height==None and n_width!=None:
w = n_width
h = np.int(1.0*batch_size/w)
elif n_height!=None and n_width!=None:
h = n_height
w = n_width
pass
image_height = np.shape(images)[1]
image_width = np.shape(images)[2]
n_channel = np.shape(images)[3]
images = np.reshape(images, [-1, image_height, image_width, n_channel])
canvas = np.empty((h * image_height, w * image_width, n_channel))
for i in range(h):
for j in range(w):
canvas[i*image_height:(i+1)*image_height, j*image_width:(j+1)*image_width, :] = \
images[h*i+j].reshape(image_height, image_width, 3)
pass
pass
plt.figure(figsize=(h, w))
plt.imshow(canvas)
label = "Epoch: {0}".format(index+1)
plt.xlabel(label)
file_name = self.plot_images_name + str(index+1)
plt.savefig(self.images_save_floder + '/' + file_name)
print(os.getcwd())
print("[OK] image saved in file: ", self.images_save_floder + '/' + file_name)
plt.show()
pass
def train(self):
'''train'''
trainset = self.load_trainset()
x = tf.placeholder(tf.float32, shape=[None, self.image_height*self.image_width*3], name="input_real")
X = tf.reshape(x, [-1] + [self.image_height, self.image_width, 3])
z = tf.placeholder(tf.float32, shape=[None, self.z_dim], name="z")
G = self.generator(z, reuse=False)
D_fake_logits = self.discriminator(G, reuse=False)
D_true_logits = self.discriminator(X, reuse=True)
G_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_fake_logits, \
labels=tf.ones_like(D_fake_logits)))
D_loss_1 = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_true_logits, \
labels=tf.ones_like(D_true_logits)))
D_loss_2 = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_fake_logits, \
labels=tf.zeros_like(D_fake_logits)))
D_loss = D_loss_1 + D_loss_2
t_vars = tf.trainable_variables()
d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
g_vars = [var for var in t_vars if var.name.startswith('generator')]
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
g_optimization = tf.train.AdamOptimizer(learning_rate=self.learning_rate, \
beta1=self.adam_beta1).minimize(G_loss, var_list=g_vars)
d_optimization = tf.train.AdamOptimizer(learning_rate=self.learning_rate, \
beta1=self.adam_beta1).minimize(D_loss, var_list=d_vars)
print("[OK] successfully make the network")
start_time = time.time()
sess = tf.Session()
sess.run(tf.initialize_all_variables())
for i in range(self.epoch):
total_batch = int(len(trainset)/self.batch_size)
d_cost = 0
g_cost = 0
for j in range(total_batch):
mini_batch = trainset[j*self.batch_size : (j*self.batch_size+self.batch_size)]
z1 = np.random.uniform(low=-1.0, high=1.0, size=[self.batch_size, self.z_dim])
d_op, d_loss = sess.run([d_optimization, D_loss], feed_dict={x: mini_batch, z: z1})
z2 = np.random.uniform(low=-1.0, high=1.0, size=[self.batch_size, self.z_dim])
g_op, g_loss = sess.run([g_optimization, G_loss], feed_dict={x: mini_batch, z: z2})
images_generated = sess.run(G, feed_dict={z: z2})
d_cost += d_loss/total_batch
g_cost += g_loss/total_batch
self.plot_save_outputs(i, images_generated)
hour = int((time.time() - start_time)/3600)
minute = int(((time.time() - start_time) - 3600*hour)/60)
sec = int((time.time() - start_time) - 3600*hour - 60*minute)
print("Time: ", hour, "h", minute, "m", sec, "s", \
"\nEpoch: ", i, "\nG_loss: ", g_cost, "\nD_loss: ", d_cost)
pass
pass
sess.close()
pass
pass
tf.reset_default_graph()
dcgan = DCGAN(dataset_floder='./faces', image_height=32, image_width=48, batch_size=49)
dcgan.train()