-
Notifications
You must be signed in to change notification settings - Fork 0
/
art_dc_gan.py
346 lines (278 loc) · 10.2 KB
/
art_dc_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
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
from __future__ import division, print_function
import argparse
import datetime
import matplotlib.pyplot as plt
import numpy as np
from keras.activations import relu
from keras.layers import (Activation, BatchNormalization, Dense, Dropout,
Flatten, Input, Reshape)
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.convolutional import Conv2D, Conv2DTranspose
from keras.models import Model, Sequential
from keras.optimizers import Adam
from image_god import ImageGod
from scrapers.wikiart_scraper import WikiartScraper
class DCGAN():
def __init__(self):
self.img_rows = 128
self.img_cols = 128
self.channels = 3
self.img_shape = (self.img_rows, self.img_cols, self.channels)
self.noise = 200
self.ngf = 64
self.ndf = 32
optimizer = Adam(0.0002, 0.5)
# Build and compile the discriminator
self.discriminator = self.build_discriminator()
self.discriminator.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
# Build and compile the generator
self.generator = self.build_generator()
self.generator.compile(loss='binary_crossentropy', optimizer=optimizer)
# The generator takes noise as input and generated imgs
z = Input((self.noise,))
img = self.generator(z)
# For the combined model we will only train the generator
self.discriminator.trainable = False
# The valid takes generated images as input and determines validity
valid = self.discriminator(img)
# The combined model (stacked generator and discriminator) takes
# noise as input => generates images => determines validity
self.combined = Model(z, valid)
self.combined.compile(loss='binary_crossentropy', optimizer=optimizer)
def build_generator(self):
k = 5
s = 2
model = Sequential()
# First Layer
model.add(Dense(4 * 4 * 1024, input_shape=(self.noise,)))
model.add(Reshape(target_shape=(4, 4, 1024)))
model.add(BatchNormalization(momentum=0.8))
model.add(Activation('relu'))
# Layer 2
model.add(Conv2DTranspose(
filters=self.ngf*8,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
))
model.add(BatchNormalization(momentum=0.8))
model.add(Activation('relu'))
# Layer 3
model.add(Conv2DTranspose(
filters=self.ngf*4,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
))
model.add(BatchNormalization(momentum=0.8))
model.add(Activation('relu'))
# Layer 4
model.add(Conv2DTranspose(
filters=self.ngf*2,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
))
model.add(BatchNormalization(momentum=0.8))
model.add(Activation('relu'))
# Layer 5
model.add(Conv2DTranspose(
filters=self.ngf,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
))
model.add(BatchNormalization(momentum=0.8))
model.add(Activation('relu'))
# Output Layer
model.add(
Conv2DTranspose(
filters=self.channels,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
)
)
model.add(Activation('tanh'))
model.summary()
noise = Input(shape=(self.noise,))
img = model(noise)
return Model(noise, img)
def build_discriminator(self):
k = 4
s = 2
model = Sequential()
# First Layer
model.add(
Conv2D(
filters=self.ndf,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
input_shape=self.img_shape,
)
)
model.add(LeakyReLU(alpha=0.2))
# Layer 2
model.add(
Conv2D(
filters=self.ndf*2,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
)
)
model.add(BatchNormalization(momentum=0.8))
model.add(LeakyReLU(alpha=0.2))
# Layer 3
model.add(
Conv2D(
filters=self.ndf*4,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
)
)
model.add(BatchNormalization(momentum=0.8))
model.add(LeakyReLU(alpha=0.2))
# Layer 4
model.add(
Conv2D(
filters=self.ndf*8,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
)
)
model.add(BatchNormalization(momentum=0.8))
model.add(LeakyReLU(alpha=0.2))
# Layer 5
model.add(
Conv2D(
filters=self.ndf*16,
kernel_size=k,
strides=s,
padding='same',
use_bias=False,
)
)
model.add(BatchNormalization(momentum=0.8))
model.add(LeakyReLU(alpha=0.2))
# Final Layer
model.add(Flatten())
model.add(Dropout(.3))
model.add(Dense(1, activation='sigmoid'))
model.summary()
img = Input(shape=self.img_shape)
validity = model(img)
return Model(img, validity)
def train(self, training_dir, epochs, batch_size=32, save_interval=100, transform=50, wikiart_scrape_url=None):
# ---------------------
# Preprocessing
# ---------------------
# Load from training_dir and normalize dataset
god = ImageGod()
# Scrape painter's collection from wikiart as Y Train if given painter
if wikiart_scrape_url:
ws = WikiartScraper()
ws.scrape_art(
wikiart_scrape_url,
training_dir,
)
god.transform_images(
self.img_shape, training_dir, epochs=transform)
training_dir = training_dir + '_transformations'
# Load from x training_dir
X_train = god.load_images(
self.img_shape,
training_dir,
)
start_time = datetime.datetime.now()
half_batch = int(batch_size / 2)
for epoch in range(epochs):
# ---------------------
# Train Discriminator
# ---------------------
# Select a random half batch of images
idx = np.random.randint(0, X_train.shape[0], half_batch)
imgs = X_train[idx]
noise = np.random.normal(0, 1, (half_batch, self.noise))
# Generate a half batch of new images
gen_imgs = self.generator.predict(noise)
# Train the discriminator
d_loss_real = self.discriminator.train_on_batch(
imgs, np.ones((half_batch, 1)))
d_loss_fake = self.discriminator.train_on_batch(
gen_imgs, np.zeros((half_batch, 1)))
d_loss = 0.5 * np.add(d_loss_real, d_loss_fake)
# ---------------------
# Train Generator
# ---------------------
noise = np.random.normal(0, 1, (batch_size, self.noise))
# The generator wants the discriminator to label the generated samples as valid (ones)
valid_y = np.ones((batch_size, 1))
# Train the generator
g_loss = self.combined.train_on_batch(noise, valid_y)
elapsed_time = datetime.datetime.now() - start_time
# Plot the progress
print("%d/%d [D loss: %f, acc.: %.2f%%] [G loss: %f] time: %s" %
(epoch, epochs, d_loss[0], 100*d_loss[1], g_loss, elapsed_time))
# If at save interval => save generated image samples
if epoch % save_interval == 0:
self.save_imgs(epoch)
# Save generator
self.generator.save('art_dc_gan_generator.h5')
def save_imgs(self, epoch):
r, c = 3, 3
noise = np.random.normal(0, 1, (r * c, self.noise))
gen_imgs = self.generator.predict(noise)
# Rescale images 0 - 1
gen_imgs = 0.5 * gen_imgs + 0.5
fig, axs = plt.subplots(r, c)
cnt = 0
for i in range(r):
for j in range(c):
axs[i, j].imshow(gen_imgs[cnt, :, :, :])
axs[i, j].axis('off')
cnt += 1
fig.savefig("samples/dcgan/art_%d.png" % epoch)
plt.close()
def parse_command_line_args():
parser = argparse.ArgumentParser(description='AI Generated Art Bitch')
parser.add_argument('epochs', type=int,
help='number of epochs')
parser.add_argument('training_dir', type=str,
help='filepath of training set (if wikiart url is given then filepath becomes the save dir)')
parser.add_argument('-b', '--batchsize',
default=32, type=int, help='size of batches per epoch')
parser.add_argument('-s', '--saveinterval',
type=int, default=100, help='interval to save sample images')
parser.add_argument('-w', '--wikiart', type=str, default=None,
help='url of wikiart profile to dowload from')
parser.add_argument('-t', '--transform', type=int, default=50,
help='number of transformations applied to each picture')
return vars(parser.parse_args())
# print(args)
if __name__ == '__main__':
wikiart_profile = 'https://www.wikiart.org/en/profile/5c9ba655edc2c9b87424edfe/albums/favourites'
args = parse_command_line_args()
gan = DCGAN()
gan.train(
training_dir=args['training_dir'],
epochs=args['epochs'],
transform=args['transform'],
batch_size=args['batchsize'],
save_interval=args['saveinterval'],
wikiart_scrape_url=args['wikiart'],
)