-
Notifications
You must be signed in to change notification settings - Fork 14
/
mnist_wacgan.py
328 lines (253 loc) · 12.3 KB
/
mnist_wacgan.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
#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
Train an Wasserstein Auxiliary Classifier Generative Adversarial Network (WACGAN) on the MNIST dataset.
See https://arxiv.org/abs/1610.09585 for more details about ACGAN.
See https://arxiv.org/abs/1701.07875 for more details about WACGAN.
You should start to see reasonable images after ~3 epochs.
According to the paper, the performance is highly related to the discriminator loss.
You should use a GPU, as the convolution-heavy operations are very slow on the CPU.
Prefer the TensorFlow backend if you plan on iterating, as the compilation time can be a blocker using Theano.
Timings:
Hardware | Backend | Time / Epoch
-------------------------------------------
CPU | TF | 3 hrs
Titan X (maxwell) | TF | 4 min
Titan X (maxwell) | TH | 7 min
Consult https://github.com/bobchennan/Wasserstein-GAN-Keras for more information and
example output
The original ACGAN implementation can be found in https://github.com/lukedeo/keras-acgan
More tricks to train GAN can be found in https://github.com/soumith/ganhacks
"""
from __future__ import print_function
from collections import defaultdict
try:
import cPickle as pickle
except ImportError:
import pickle
from PIL import Image
from six.moves import range
import keras.backend as K
from keras.datasets import mnist
from keras.layers import Input, Dense, Reshape, Flatten, Embedding, merge, Dropout
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.convolutional import UpSampling2D, Convolution2D
from keras.layers.noise import GaussianNoise
from keras.models import Sequential, Model
from keras.optimizers import Adam, SGD
from keras.backend.common import _EPSILON
from keras.utils.generic_utils import Progbar
import numpy as np
np.random.seed(1331)
K.set_image_dim_ordering('th')
def modified_binary_crossentropy(target, output):
#output = K.clip(output, _EPSILON, 1.0 - _EPSILON)
#return -(target * output + (1.0 - target) * (1.0 - output))
return K.mean(target*output)
def build_generator(latent_size):
# we will map a pair of (z, L), where z is a latent vector and L is a
# label drawn from P_c, to image space (..., 1, 28, 28)
cnn = Sequential()
cnn.add(Dense(1024, input_dim=latent_size))
cnn.add(LeakyReLU())
cnn.add(Dense(128 * 7 * 7))
cnn.add(LeakyReLU())
cnn.add(Reshape((128, 7, 7)))
# upsample to (..., 14, 14)
cnn.add(UpSampling2D(size=(2, 2)))
cnn.add(Convolution2D(256, 5, 5, border_mode='same',
init='glorot_uniform'))
cnn.add(LeakyReLU())
# upsample to (..., 28, 28)
cnn.add(UpSampling2D(size=(2, 2)))
cnn.add(Convolution2D(128, 5, 5, border_mode='same',
init='glorot_uniform'))
cnn.add(LeakyReLU())
# take a channel axis reduction
cnn.add(Convolution2D(1, 2, 2, border_mode='same',
activation='tanh', init='glorot_uniform'))
# this is the z space commonly refered to in GAN papers
latent = Input(shape=(latent_size, ))
# this will be our label
image_class = Input(shape=(1,), dtype='int32')
# 10 classes in MNIST
cls = Flatten()(Embedding(10, latent_size,
init='glorot_uniform')(image_class))
# hadamard product between z-space and a class conditional embedding
h = merge([latent, cls], mode='mul')
fake_image = cnn(h)
return Model(input=[latent, image_class], output=fake_image)
def build_discriminator():
# build a relatively standard conv net, with LeakyReLUs as suggested in
# the reference paper
cnn = Sequential()
#cnn.add(GaussianNoise(0.2, input_shape=(1, 28, 28)))
cnn.add(Convolution2D(32, 3, 3, border_mode='same', subsample=(2, 2),
input_shape=(1, 28, 28)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(64, 3, 3, border_mode='same', subsample=(1, 1)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(128, 3, 3, border_mode='same', subsample=(2, 2)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Convolution2D(256, 3, 3, border_mode='same', subsample=(1, 1)))
cnn.add(LeakyReLU())
cnn.add(Dropout(0.3))
cnn.add(Flatten())
image = Input(shape=(1, 28, 28))
features = cnn(image)
# first output (name=generation) is whether or not the discriminator
# thinks the image that is being shown is fake, and the second output
# (name=auxiliary) is the class that the discriminator thinks the image
# belongs to.
fake = Dense(1, activation='linear', name='generation')(features)
aux = Dense(10, activation='softmax', name='auxiliary')(features)
return Model(input=image, output=[fake, aux])
if __name__ == '__main__':
# batch and latent size taken from the paper
nb_epochs = 50
batch_size = 100
latent_size = 100
# Adam parameters suggested in https://arxiv.org/abs/1511.06434
adam_lr = 0.0002
adam_beta_1 = 0.5
# build the discriminator
discriminator = build_discriminator()
discriminator.compile(
optimizer=SGD(clipvalue=0.01),#Adam(lr=adam_lr, beta_1=adam_beta_1),
loss=[modified_binary_crossentropy, 'sparse_categorical_crossentropy']
)
# build the generator
generator = build_generator(latent_size)
generator.compile(optimizer=Adam(lr=adam_lr, beta_1=adam_beta_1),
loss='binary_crossentropy')
latent = Input(shape=(latent_size, ))
image_class = Input(shape=(1,), dtype='int32')
# get a fake image
fake = generator([latent, image_class])
# we only want to be able to train generation for the combined model
discriminator.trainable = False
fake, aux = discriminator(fake)
combined = Model(input=[latent, image_class], output=[fake, aux])
combined.compile(
optimizer='RMSprop',
loss=[modified_binary_crossentropy, 'sparse_categorical_crossentropy']
)
# get our mnist data, and force it to be of shape (..., 1, 28, 28) with
# range [-1, 1]
(X_train, y_train), (X_test, y_test) = mnist.load_data()
X_train = (X_train.astype(np.float32) - 127.5) / 127.5
X_train = np.expand_dims(X_train, axis=1)
X_test = (X_test.astype(np.float32) - 127.5) / 127.5
X_test = np.expand_dims(X_test, axis=1)
nb_train, nb_test = X_train.shape[0], X_test.shape[0]
train_history = defaultdict(list)
test_history = defaultdict(list)
for epoch in range(nb_epochs):
print('Epoch {} of {}'.format(epoch + 1, nb_epochs))
nb_batches = int(X_train.shape[0] / batch_size)
progress_bar = Progbar(target=nb_batches)
epoch_gen_loss = []
epoch_disc_loss = []
for index in range(nb_batches):
if len(epoch_gen_loss) + len(epoch_disc_loss) > 1:
progress_bar.update(index, values=[('disc_loss',np.mean(np.array(epoch_disc_loss),axis=0)[0]), ('gen_loss', np.mean(np.array(epoch_gen_loss),axis=0)[0])])
else:
progress_bar.update(index)
# generate a new batch of noise
#noise = np.random.uniform(-1, 1, (batch_size, latent_size))
noise = np.random.normal(0, 1, (batch_size, latent_size))
# get a batch of real images
image_batch = X_train[index * batch_size:(index + 1) * batch_size]
label_batch = y_train[index * batch_size:(index + 1) * batch_size]
# sample some labels from p_c
sampled_labels = np.random.randint(0, 10, batch_size)
# generate a batch of fake images, using the generated labels as a
# conditioner. We reshape the sampled labels to be
# (batch_size, 1) so that we can feed them into the embedding
# layer as a length one sequence
generated_images = generator.predict(
[noise, sampled_labels.reshape((-1, 1))], verbose=0)
X = np.concatenate((image_batch, generated_images))
y = np.array([-1] * batch_size + [1] * batch_size)
aux_y = np.concatenate((label_batch, sampled_labels), axis=0)
# see if the discriminator can figure itself out...
epoch_disc_loss.append(discriminator.train_on_batch(X, [y, aux_y]))
# make new noise. we generate 2 * batch size here such that we have
# the generator optimize over an identical number of images as the
# discriminator
#noise = np.random.uniform(-1, 1, (2 * batch_size, latent_size))
noise = np.random.normal(0, 1, (2 * batch_size, latent_size))
sampled_labels = np.random.randint(0, 10, 2 * batch_size)
# we want to train the genrator to trick the discriminator
# For the generator, we want all the {fake, not-fake} labels to say
# not-fake
trick = -np.ones(2 * batch_size)
epoch_gen_loss.append(combined.train_on_batch(
[noise, sampled_labels.reshape((-1, 1))], [trick, sampled_labels]))
print('\nTesting for epoch {}:'.format(epoch + 1))
# evaluate the testing loss here
# generate a new batch of noise
#noise = np.random.uniform(-1, 1, (nb_test, latent_size))
noise = np.random.normal(0, 1, (nb_test, latent_size))
# sample some labels from p_c and generate images from them
sampled_labels = np.random.randint(0, 10, nb_test)
generated_images = generator.predict(
[noise, sampled_labels.reshape((-1, 1))], verbose=False)
X = np.concatenate((X_test, generated_images))
y = np.array([1] * nb_test + [0] * nb_test)
aux_y = np.concatenate((y_test, sampled_labels), axis=0)
# see if the discriminator can figure itself out...
discriminator_test_loss = discriminator.evaluate(
X, [y, aux_y], verbose=False)
discriminator_train_loss = np.mean(np.array(epoch_disc_loss), axis=0)
# make new noise
#noise = np.random.uniform(-1, 1, (2 * nb_test, latent_size))
noise = np.random.normal(0, 1, (2 * nb_test, latent_size))
sampled_labels = np.random.randint(0, 10, 2 * nb_test)
trick = np.ones(2 * nb_test)
generator_test_loss = combined.evaluate(
[noise, sampled_labels.reshape((-1, 1))],
[trick, sampled_labels], verbose=False)
generator_train_loss = np.mean(np.array(epoch_gen_loss), axis=0)
# generate an epoch report on performance
train_history['generator'].append(generator_train_loss)
train_history['discriminator'].append(discriminator_train_loss)
test_history['generator'].append(generator_test_loss)
test_history['discriminator'].append(discriminator_test_loss)
print('{0:<22s} | {1:4s} | {2:15s} | {3:5s}'.format(
'component', *discriminator.metrics_names))
print('-' * 65)
ROW_FMT = '{0:<22s} | {1:<4.2f} | {2:<15.2f} | {3:<5.2f}'
print(ROW_FMT.format('generator (train)',
*train_history['generator'][-1]))
print(ROW_FMT.format('generator (test)',
*test_history['generator'][-1]))
print(ROW_FMT.format('discriminator (train)',
*train_history['discriminator'][-1]))
print(ROW_FMT.format('discriminator (test)',
*test_history['discriminator'][-1]))
# save weights every epoch
generator.save_weights(
'params_generator_epoch_{0:03d}.hdf5'.format(epoch), True)
discriminator.save_weights(
'params_discriminator_epoch_{0:03d}.hdf5'.format(epoch), True)
# generate some digits to display
#noise = np.random.uniform(-1, 1, (100, latent_size))
noise = np.random.normal(-1, 1, (100, latent_size))
sampled_labels = np.array([
[i] * 10 for i in range(10)
]).reshape(-1, 1)
# get a batch to display
generated_images = generator.predict(
[noise, sampled_labels], verbose=0)
# arrange them into a grid
img = (np.concatenate([r.reshape(-1, 28)
for r in np.split(generated_images, 10)
], axis=-1) * 127.5 + 127.5).astype(np.uint8)
Image.fromarray(img).save(
'plot_epoch_{0:03d}_generated.png'.format(epoch))
pickle.dump({'train': train_history, 'test': test_history},
open('acgan-history.pkl', 'wb'))