-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
133 lines (108 loc) · 4.57 KB
/
main.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
import numpy as np
from tqdm import tqdm
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from keras import initializers
from keras.models import Model, Sequential
from keras.layers import Input, Dense, Conv2D, MaxPool2D, Flatten, Dropout
from keras.layers.advanced_activations import LeakyReLU
from keras.optimizers import Adam
np.random.seed(0)
class CatFacesGAN:
def __init__(self, learning_rate, beta, summaries):
self._learning_rate = learning_rate
self._beta = beta
self._show_summaries = summaries
self._random_input_dim = 100 # tensor dimenzije
def load_generator(self):
G = Sequential() # keras model
G.add(Dense(256, input_dim=self._random_input_dim, kernel_initializer=initializers.RandomNormal(stddev=0.02)))
G.add(LeakyReLU(0.3)) # activation layer
G.add(Dense(512))
G.add(LeakyReLU(0.3))
G.add(Dense(1024))
G.add(LeakyReLU(0.2))
G.add(Dense(2048))
G.add(LeakyReLU(0.2))
G.add(Dense(784, activation='tanh')) # output mu je izmedju 0 ili 1
# binary_crossentropy - izlaz je 0 ili 1
G.compile(loss='binary_crossentropy', optimizer=Adam(lr=self._learning_rate, beta_1=self._beta))
if self._show_summaries:
G.summary()
return G
def load_discriminator(self):
D = Sequential()
D.add(Dense(1024, input_dim=784, kernel_initializer=initializers.RandomNormal(stddev=0.02)))
D.add(LeakyReLU(0.2))
D.add(Dropout(0.3))
D.add(Dense(512))
D.add(LeakyReLU(0.2))
D.add(Dropout(0.3))
D.add(Dense(256))
D.add(LeakyReLU(0.2))
D.add(Dropout(0.3))
D.add(Dense(1, activation='sigmoid'))
D.compile(loss='binary_crossentropy', optimizer=Adam(lr=self._learning_rate, beta_1=self._beta))
if self._show_summaries:
D.summary()
return D
def load_GAN(self, discriminator, generator, random_input_dim):
discriminator.trainable = False # stavimo kako bi sacuvali pocetno stanje mreze
catgan_input = Input(shape=[self._random_input_dim, ])
x = generator(catgan_input)
catgan_output = discriminator(x)
catgan = Model(inputs=catgan_input, outputs=catgan_output)
catgan.compile(loss='binary_crossentropy', optimizer=Adam(lr=self._learning_rate, beta_1=self._beta))
catgan.summary()
return catgan
def save_generated_images(self, epoch, generator, show_images=False):
num_examples = 100
random_noise = np.random.normal(0, 1, size=[num_examples, self._random_input_dim])
generated_images = generator.predict(random_noise)
generated_images = generated_images.reshape(num_examples, 28, 28)
plt.figure(figsize=(10, 10))
for i in range(generated_images.shape[0]):
plt.subplot(10, 10, i + 1)
plt.imshow(generated_images[i], cmap='gray_r')
plt.axis('off')
plt.tight_layout()
plt.savefig('generisana_slika_epoha_{}.png'.format(epoch))
if show_images:
plt.show()
def train_gan(self, x_train, num_epochs, batch_size=256):
batch_count = int(x_train.shape[0] // batch_size)
generator = self.load_generator()
discriminator = self.load_discriminator()
catgan = self.load_GAN(discriminator, generator, self._random_input_dim)
for epoch in range(1, num_epochs + 1):
for i in tqdm(range(batch_count), desc="Epoch {}".format(epoch)):
random_noise = np.random.normal(0, 1, size=[batch_size, self._random_input_dim])
image_batch = x_train[np.random.randint(0, x_train.shape[0], size=batch_size)]
generated_images = generator.predict(random_noise)
X_discriminator = np.concatenate([image_batch, generated_images])
y_discriminator = np.zeros(2 * batch_size)
y_discriminator[:batch_size] = 0.90
discriminator.trainable = True
discriminator.train_on_batch(X_discriminator, y_discriminator)
random_noise = np.random.normal(0, 1, size=[batch_size, self._random_input_dim])
y_generator = np.ones(batch_size)
discriminator.trainable = False
catgan.train_on_batch(random_noise, y_generator)
if epoch == 1 or epoch % 20 == 0:
self.save_generated_images(epoch, generator)
def load_cats(self):
try:
cats = np.load("./full-numpy_bitmap-cat.npy")
Y = []
for i in range(cats.shape[0]): # uzimamo prvu dimenziju aka sirina
Y.append([1, 0])
Y = np.array(Y) # niz za outputs mreze
(x_train, y_train, x_test, y_test) = train_test_split(cats, Y)
x_train = (x_train.astype(np.float32)) / 255 # moramo da skaliramo rgb vrednosti od 0 do 1
x_train = x_train.reshape(x_train.shape[0], 784) # kreiramo ulaznu sliku od 784px
return (x_train, y_train, x_test, y_test)
except Exception as e:
print(e)
catgan = CatFacesGAN(learning_rate=0.0002, beta=0.5, summaries=False)
(x_train, y_train, x_test, y_test) = catgan.load_cats()
catgan.train_gan(x_train, 200, 256)