-
Notifications
You must be signed in to change notification settings - Fork 1
/
train.py
424 lines (307 loc) · 16.5 KB
/
train.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
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
import os, time, itertools, imageio, pickle
import numpy as np
import matplotlib.pyplot as plt
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data
import tensorflow.contrib.slim as slim
sess = tf.InteractiveSession()
# load MNIST
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True, reshape=[])
imgs=mnist.train.images
labels=mnist.train.labels
# MNIST resize and normalization
train_set_imgs = tf.image.resize_images(imgs, [64, 64]).eval()
train_set_imgs = (train_set_imgs - 0.5) / 0.5 # normalization; range: -1 ~ 1
train_set_txt = (labels - 0.5)/0.5
# training parameters
batch_size = 100
lr_d = 0.0003
lr_g =0.0003
#lr_g2= 0.00002
train_epoch = 200
vocab=['0','1','2','3','4','5','6','7','8','9']
def lrelu(x, th=0.2):
return tf.maximum(th * x, x)
def KL_loss_txt(mu,sigma,reuse=False):
with tf.variable_scope("generator_img_KL_divergence",reuse=reuse):
#loss=tf.distributions.kl_divergence(mu,sigma,allow_nan_stats=False)
loss=-sigma + .5 * (-1 + tf.exp(2. * sigma) + tf.square(mu))
loss = tf.reduce_mean(loss)
return loss
def KL_loss_img(mu,sigma,reuse=False):
with tf.variable_scope("generator_txt_KL_divergence",reuse=reuse):
#loss=tf.distributions.kl_divergence(mu,sigma,allow_nan_stats=False)
loss=-sigma + .5 * (-1 + tf.exp(2. * sigma) + tf.square(mu))
loss = tf.reduce_mean(loss)
return loss
def generate_cond_img(txt,reuse=False):
with tf.variable_scope("generator_img_cond",reuse=reuse):
f=tf.layers.Flatten()(txt)
fc=tf.layers.dense(f,20,use_bias=True,bias_initializer=tf.zeros_initializer(),trainable=True,name=None,reuse=reuse)
lr=lrelu(fc,0.2)
mean=lr[:,:10]
stddev=lr[:,10:]
dist=tf.truncated_normal(tf.shape(mean))
c_out=mean+ stddev*dist
return c_out,mean,stddev
def generate_cond_txt(img,reuse=False):
with tf.variable_scope("generator_txt_cond",reuse=reuse):
f=tf.layers.Flatten()(img)
fc=tf.layers.dense(f,2*64*64,use_bias=True,bias_initializer=tf.zeros_initializer(),trainable=True,name=None,reuse=reuse)
lr=lrelu(fc,0.2)
mean=lr[:,:(64*64)]
stddev=lr[:,(64*64):]
dist=tf.truncated_normal(tf.shape(mean))
c_out=mean+ stddev*dist
return c_out,mean,stddev
def show_all_variables():
model_vars = tf.trainable_variables()
slim.model_analyzer.analyze_vars(model_vars, print_info=True)
def decode_embed(array, vocab):
array=list(array)
index=array.index(max(array))
return vocab[index]
def generator(x, isTrain=True, reuse=False):
with tf.variable_scope('generator', reuse=reuse):
conv1 = tf.layers.conv2d_transpose(x, 512, [4, 4], strides=(2, 2), padding='valid')
lrelu1 = lrelu(tf.layers.batch_normalization(conv1, training=isTrain), 0.2)
conv2 = tf.layers.conv2d_transpose(lrelu1, 512, [3, 3], strides=(1, 1), padding='same')
lrelu2 = lrelu(tf.layers.batch_normalization(conv2, training=isTrain), 0.2)
conv3 = tf.layers.conv2d_transpose(lrelu2, 128, [3, 3], strides=(1, 1), padding='same')
lrelu3 = lrelu(tf.layers.batch_normalization(conv3, training=isTrain), 0.2)
conv4 = tf.layers.conv2d_transpose(lrelu3, 128, [1, 1], strides=(1, 1), padding='same')
lrelu4 = lrelu(tf.layers.batch_normalization(conv4, training=isTrain), 0.2)
conv5 = tf.layers.conv2d_transpose(lrelu4, 512, [4, 4], strides=(2, 2), padding='same')
lrelu5 = lrelu(tf.layers.batch_normalization(conv4, training=isTrain), 0.2)
conv6 = tf.layers.conv2d_transpose(lrelu5, 256, [4, 4], strides=(2, 2), padding='same')
lrelu6 = lrelu(tf.layers.batch_normalization(conv6, training=isTrain), 0.2)
conv7 = tf.layers.conv2d_transpose(lrelu6, 128, [4, 4], strides=(2, 2), padding='same')
lrelu7 = lrelu(tf.layers.batch_normalization(conv7, training=isTrain), 0.2)
conv8 = tf.layers.conv2d_transpose(lrelu7, 64, [4, 4], strides=(2, 2), padding='same')
lrelu8 = lrelu(tf.layers.batch_normalization(conv8, training=isTrain), 0.2)
conv9 = tf.layers.conv2d_transpose(lrelu8, 1, [4, 4], strides=(2, 2), padding='same')
o_img = tf.nn.tanh(conv9)
lrelu_txt=tf.layers.Flatten()(lrelu(lrelu1,0.2))
op_txt= tf.layers.dense(lrelu_txt,10,use_bias=True,bias_initializer=tf.zeros_initializer(),trainable=True,name=None,reuse=reuse)
o_txt = tf.nn.tanh(op_txt)
return o_img,o_txt
# D(x)
def discriminator_combined(x,y, isTrain=True, reuse=False):
with tf.variable_scope('discriminator_combined', reuse=reuse):
conv1 = tf.layers.conv2d(x, 64, [4, 4], strides=(2, 2), padding='same')
lrelu1 = lrelu(conv1, 0.2)
conv2 = tf.layers.conv2d(lrelu1, 128, [4, 4], strides=(2, 2), padding='same')
lrelu2 = lrelu(tf.layers.batch_normalization(conv2, training=isTrain), 0.2)
conv3 = tf.layers.conv2d(lrelu2, 256, [4, 4], strides=(2, 2), padding='same')
lrelu3 = lrelu(tf.layers.batch_normalization(conv3, training=isTrain), 0.2)
conv4 = tf.layers.conv2d(lrelu3, 512, [4, 4], strides=(2, 2), padding='same')
lrelu4 = lrelu(tf.layers.batch_normalization(conv4, training=isTrain), 0.2)
conv5 = tf.layers.conv2d(lrelu4, 128, [1, 1], strides=(1, 1), padding='same')
lrelu5= lrelu(tf.layers.batch_normalization(conv5, training=isTrain), 0.2)
conv6 = tf.layers.conv2d(lrelu5, 128, [3, 3], strides=(1, 1), padding='same')
lrelu6= lrelu(tf.layers.batch_normalization(conv6, training=isTrain), 0.2)
conv7 = tf.layers.conv2d(lrelu6, 512, [3, 3], strides=(1, 1), padding='same')
lrelu7= lrelu(tf.layers.batch_normalization(conv7, training=isTrain), 0.2)
fc= tf.layers.dense(y,10,use_bias=True,bias_initializer=tf.zeros_initializer(),trainable=True,name=None,reuse=reuse)
lrelu_fc=lrelu(fc,0.2)
z=tf.expand_dims(tf.expand_dims(lrelu_fc,1),1)
txt=tf.tile(z,[1,4,4,1])
img=tf.concat([lrelu7,txt],axis=3)
conv8 = tf.layers.conv2d(img, 512, [1, 1], strides=(1, 1), padding='same')
lrelu8= lrelu(tf.layers.batch_normalization(conv8, training=isTrain), 0.2)
o = tf.layers.conv2d(lrelu8, 1, [4, 4], strides=(4, 4), padding='valid')
return o
def discriminator_img(x,isTrain=True, reuse=False):
with tf.variable_scope('discriminator_img', reuse=reuse):
conv1 = tf.layers.conv2d(x, 64, [4, 4], strides=(2, 2), padding='same')
lrelu1 = lrelu(conv1, 0.2)
conv2 = tf.layers.conv2d(lrelu1, 128, [4, 4], strides=(2, 2), padding='same')
lrelu2 = lrelu(tf.layers.batch_normalization(conv2, training=isTrain), 0.2)
conv3 = tf.layers.conv2d(lrelu2, 256, [4, 4], strides=(2, 2), padding='same')
lrelu3 = lrelu(tf.layers.batch_normalization(conv3, training=isTrain), 0.2)
conv4 = tf.layers.conv2d(lrelu3, 512, [4, 4], strides=(2, 2), padding='same')
lrelu4 = lrelu(tf.layers.batch_normalization(conv4, training=isTrain), 0.2)
conv5 = tf.layers.conv2d(lrelu4, 128, [1, 1], strides=(1, 1), padding='same')
lrelu5= lrelu(tf.layers.batch_normalization(conv5, training=isTrain), 0.2)
conv6 = tf.layers.conv2d(lrelu5, 128, [3, 3], strides=(1, 1), padding='same')
lrelu6= lrelu(tf.layers.batch_normalization(conv6, training=isTrain), 0.2)
conv7 = tf.layers.conv2d(lrelu6, 512, [3, 3], strides=(1, 1), padding='same')
lrelu7= lrelu(tf.layers.batch_normalization(conv7, training=isTrain), 0.2)
conv8 = tf.layers.conv2d(lrelu7, 512, [1, 1], strides=(1, 1), padding='same')
lrelu8= lrelu(tf.layers.batch_normalization(conv8, training=isTrain), 0.2)
o = tf.layers.conv2d(lrelu8, 1, [4, 4], strides=(4, 4), padding='valid')
return o
def discriminator_txt(y, isTrain=True, reuse=False):
with tf.variable_scope('discriminator_txt', reuse=reuse):
fc= tf.layers.dense(y,10,use_bias=True,bias_initializer=tf.zeros_initializer(),trainable=True,name=None,reuse=reuse)
lrelu_fc=lrelu(fc,0.2)
z=tf.expand_dims(tf.expand_dims(lrelu_fc,1),1)
txt=tf.tile(z,[1,4,4,1])
conv8 = tf.layers.conv2d(txt, 512, [1, 1], strides=(1, 1), padding='same')
lrelu8= lrelu(tf.layers.batch_normalization(conv8, training=isTrain), 0.2)
o = tf.layers.conv2d(lrelu8, 1, [4, 4], strides=(4, 4), padding='valid')
return o
tsi=train_set_imgs[:batch_size]
tst=train_set_txt[: batch_size]
fixed_z_=np.random.normal(0, 1, (batch_size,100))
def show_result(num_epoch, show = False, save = False, path = 'result.png'):
test_images = sess.run(G_z_img, {z: fixed_z_,x:tsi,y:tst, isTrain: False})
test_texts = sess.run(G_z_txt, {z: fixed_z_,x:tsi,y:tst, isTrain: False})
size_figure_grid = 10
fig, ax = plt.subplots(size_figure_grid, size_figure_grid, figsize=(10, 10))
for i, j in itertools.product(range(size_figure_grid), range(size_figure_grid)):
ax[i, j].get_xaxis().set_visible(False)
ax[i, j].get_yaxis().set_visible(False)
for k in range(size_figure_grid*size_figure_grid):
i = k // size_figure_grid
j = k % size_figure_grid
ax[i, j].cla()
ax[i, j].imshow(np.reshape(test_images[k], (64, 64)), cmap='gray')
label = 'Epoch {0}'.format(num_epoch)
fig.text(0.5, 0.04, label, ha='center')
if save:
plt.savefig(path)
if show:
plt.show()
else:
plt.close()
gen_num=[]
for i in range(100):
gen_num.append(decode_embed(test_texts[i,:],vocab))
f=open('Results/generated_text','a+')
f.write('epoch{},Generated Nos.{}\n'.format(epoch,gen_num))
f.close()
def show_train_hist(hist, show = False, save = False, path = 'Train_hist.png'):
x = range(len(hist['D_losses']))
y1 = hist['D_losses']
y2 = hist['G_losses']
y3 = hist['D_losses_dash']
plt.plot(x, y1, label='D_loss')
plt.plot(x, y2, label='G_loss')
plt.plot(x, y3,label='D_loss_dash')
plt.xlabel('Epoch')
plt.ylabel('Loss')
plt.legend(loc=4)
plt.grid(True)
plt.tight_layout()
if save:
plt.savefig(path)
if show:
plt.show()
else:
plt.close()
# variables : input
x = tf.placeholder(tf.float32, shape=(batch_size, 64, 64, 1))
#x_wrong=tf.placeholder(tf.float32, shape=(batch_size, 64, 64, 1))
y=tf.placeholder(tf.float32,shape=(batch_size,10))
#y_wrong=tf.placeholder(tf.float32,shape=(batch_size,10))
z = tf.placeholder(tf.float32, shape=(batch_size, 100))
isTrain = tf.placeholder(dtype=tf.bool)
img_cond, mu1,stddev1=generate_cond_img(y)
txt_cond, mu2,stddev2=generate_cond_txt(x)
kl_loss_img=KL_loss_txt(mu1,stddev1)
kl_loss_txt=KL_loss_img(mu2,stddev2)
z_train=tf.expand_dims(tf.expand_dims(tf.concat([img_cond,txt_cond,z],1),1),1)
#z_txt=tf.expand_dims(tf.expand_dims(tf.concat([txt_cond,z],1),1),1)
# networks : generator
G_z_img,G_z_txt=generator(z_train, isTrain)
#G_z_txt = generator_txt(z_train, isTrain)
# networks : discriminator
D_real_logits = discriminator_combined(x,y, isTrain)
#D_wrong_logits = discriminator(x_wrong,y_wrong,isTrain,reuse=True)
D_fake_logits = discriminator_combined(G_z_img, G_z_txt, isTrain, reuse=True)
D_img_logits_real=discriminator_img(x,isTrain)
D_txt_logits_real=discriminator_txt(y,isTrain)
D_img_logits_fake=discriminator_img(G_z_img,isTrain,reuse=True)
D_txt_logits_fake=discriminator_txt(G_z_txt,isTrain,reuse=True)
# loss for each network
D_loss_real = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_real_logits, labels=tf.ones_like(D_real_logits)))
#D_loss_wrong= tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_wrong_logits, labels=tf.zeros_like(D_wrong_logits)))
D_loss_fake = 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_real + D_loss_fake #+ D_loss_wrong)/2
G_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_fake_logits, labels=tf.ones_like(D_fake_logits)))+kl_loss_img+kl_loss_txt
#G_txt_loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_fake_logits, labels=tf.ones_like(D_fake_logits)))+kl_loss_txt
D_img_loss_real=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_img_logits_real, labels=tf.ones_like(D_img_logits_real)))
D_txt_loss_real=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_txt_logits_real, labels=tf.ones_like(D_txt_logits_real)))
D_img_loss_fake=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_img_logits_fake, labels=tf.zeros_like(D_img_logits_fake)))
D_txt_loss_fake=tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=D_txt_logits_fake, labels=tf.zeros_like(D_txt_logits_fake)))
D_img_loss=D_img_loss_real+D_img_logits_fake
D_txt_loss=D_txt_loss_real+D_txt_loss_fake
D_loss_dash=tf.reduce_mean(D_loss-(D_img_loss+D_txt_loss))
# trainable variables for each network
T_vars = tf.trainable_variables()
D_vars = [var for var in T_vars if var.name.startswith('discriminator')]
#D_img_vars = [var for var in T_vars if var.name.startswith('discriminator_img')]
#D_txt_vars = [var for var in T_vars if var.name.startswith('discriminator_txt')]
G_vars = [var for var in T_vars if var.name.startswith('generator')]
#G_txt_vars = [var for var in T_vars if var.name.startswith('generator_txt')]
# optimizer for each network
with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
D_optim = tf.train.AdamOptimizer(lr_d, beta1=0.5).minimize(D_loss, var_list=D_vars)
G_optim = tf.train.AdamOptimizer(lr_g, beta1=0.5).minimize(G_loss, var_list=G_vars)
#G_txt_optim = tf.train.AdamOptimizer(lr_g2, beta1=0.5).minimize(G_txt_loss, var_list=G_txt_vars)
show_all_variables()
# open session and initialize all variables
tf.global_variables_initializer().run()
saver = tf.train.Saver(tf.global_variables())
# results save folder
root = 'MNIST_results/'
model = 'MNIST_'
if not os.path.isdir(root):
os.mkdir(root)
if not os.path.isdir(root + 'Fixed_results'):
os.mkdir(root + 'Fixed_results')
train_hist = {}
train_hist['D_losses'] = []
train_hist['G_losses'] = []
train_hist['D_losses_dash'] = []
train_hist['per_epoch_ptimes'] = []
train_hist['total_ptime'] = []
# training-loop
np.random.seed(int(time.time()))
print('training start!')
start_time = time.time()
for epoch in range(train_epoch):
G_losses = []
D_losses_dash=[]
D_losses = []
epoch_start_time = time.time()
iter=0
for iter in range(mnist.train.num_examples // batch_size):
# update discriminator
x_ = train_set_imgs[iter*batch_size:(iter+1)*batch_size]
y_ = train_set_txt[iter*batch_size:(iter+1)*batch_size]
z_ = np.random.normal(0, 1, (batch_size,100))
loss_d_, _,d_dash = sess.run([D_loss, D_optim,D_loss_dash], {x: x_,y:y_,z:z_ ,isTrain: True})
D_losses.append(loss_d_)
D_losses_dash.append(d_dash)
# update generators
loss_g, _ ,img,txt= sess.run([G_loss, G_optim,G_z_img,G_z_txt], {z: z_, x: x_,y: y_, isTrain: True})
G_losses.append(loss_g)
#loss_g_txt, _,txt = sess.run([G_txt_loss, G_txt_optim,G_z_txt], {z: z_,x: x_, y: y_, isTrain: True})
#G_txt_losses.append(loss_g_txt)
epoch_end_time = time.time()
per_epoch_ptime = epoch_end_time - epoch_start_time
print('[%d/%d] - ptime: %.2f loss_d: %.8f, loss_g: %.8f, loss_ddash: %.8f' % ((epoch + 1), train_epoch, per_epoch_ptime,
np.mean(D_losses), np.mean(G_losses),np.mean(D_losses_dash)))
fixed_p = root + 'Fixed_results/' + model + str(epoch + 1) + '.png'
show_result((epoch + 1), save=True, path=fixed_p)
train_hist['D_losses'].append(np.mean(D_losses))
train_hist['G_losses'].append(np.mean(G_losses))
train_hist['D_losses_dash'].append(np.mean(D_losses_dash))
train_hist['per_epoch_ptimes'].append(per_epoch_ptime)
saver.save(sess, "saved/model.ckpt")
end_time = time.time()
total_ptime = end_time - start_time
train_hist['total_ptime'].append(total_ptime)
print('Avg per epoch ptime: %.2f, total %d epochs ptime: %.2f' % (np.mean(train_hist['per_epoch_ptimes']), train_epoch, total_ptime))
print("Training finish!... save training results")
with open(root + model + 'train_hist.pkl', 'wb') as f:
pickle.dump(train_hist, f)
show_train_hist(train_hist, save=True, path=root + model + 'train_hist.png')
images = []
for e in range(train_epoch):
img_name = root + 'img_results/' + model + str(e + 1) + '.png'
images.append(imageio.imread(img_name))
imageio.mimsave(root + model + 'generation_animation.gif', images, fps=5)
sess.close()