-
Notifications
You must be signed in to change notification settings - Fork 4
/
cgan.py
151 lines (117 loc) · 6.25 KB
/
cgan.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
import scipy
import datetime
import matplotlib.pyplot as plt
import sys
from loader import DataLoader
import numpy as np
import os
from keras.datasets import mnist
from keras_contrib.layers.normalization.instancenormalization import InstanceNormalization
from keras.layers import Input, Dense, Reshape, Flatten, Dropout, Concatenate
from keras.layers import BatchNormalization, Activation, ZeroPadding2D
from keras.layers.advanced_activations import LeakyReLU
from keras.layers.convolutional import UpSampling2D, Conv2D
from keras.models import Sequential, Model
from keras.optimizers import Adam
from models.refiner import build_refiner
from models.classifier import build_classifier
from models.discriminator import build_discriminator, build_feature_discriminator
from models.encoder import build_encoder
class CGAN():
def __init__(self):
self.img_rows = 128
self.img_cols = 128
self.channels = 3
self.n_features = 128
self.n_classes = 31
self.img_shape = (self.img_rows, self.img_cols, self.channels)
self.data_loader = DataLoader(img_res=(self.img_rows, self.img_cols), n_classes=self.n_classes)
optimizer = Adam(0.0002, 0.5)
self.D_R = build_discriminator(self.img_shape)
self.D_F = build_feature_discriminator(self.n_features)
self.D_R.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
self.D_F.compile(loss='binary_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
self.Refiner = build_refiner(self.img_shape, self.channels)
self.Feature = build_encoder(self.img_shape, self.n_features)
self.Classifier = build_classifier(self.n_features, self.n_classes)
self.D_R.trainable = False
self.D_F.trainable = False
self.Classifier.compile(loss='categorical_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
self.Classifier.trainable = False
self.GAN_1 = Sequential()
self.GAN_1.add(self.Refiner)
self.GAN_1.add(self.D_R)
self.GAN_1.compile(loss='binary_crossentropy', optimizer=optimizer, metrics=['accuracy'])
self.GAN_2 = Sequential()
self.GAN_2.add(self.Refiner)
self.GAN_2.add(self.Feature)
self.GAN_2.add(self.D_F)
self.GAN_2.compile(loss='binary_crossentropy', optimizer=optimizer, metrics=['accuracy'])
self.GAN_3 = Sequential()
self.GAN_3.add(self.Refiner)
self.GAN_3.add(self.Feature)
self.GAN_3.add(self.Classifier)
self.GAN_3.compile(loss='categorical_crossentropy', optimizer=optimizer, metrics=['accuracy'])
def train(self, epochs, batch_size=1, interval=50):
start_time = datetime.datetime.now()
valid = np.ones((batch_size,))
refined = np.zeros((batch_size,))
for epoch in range(epochs):
for batch_i, (imgs_sim, imgs_target, classes) in enumerate(self.data_loader.load_batch(batch_size)):
imgs_refined = self.Refiner.predict(imgs_sim)
feature_sim = self.Feature.predict(imgs_sim)
feature_target = self.Feature.predict(imgs_target)
feature_refined = self.Feature.predict(imgs_refined)
dimg_loss_real = self.D_R.train_on_batch(imgs_target, valid)
dimg_loss_refined = self.D_R.train_on_batch(imgs_refined, refined)
dimg_loss = 0.5 * np.add(dimg_loss_real, dimg_loss_refined)
dfeature_loss_real = self.D_F.train_on_batch(feature_target, valid)
dfeature_loss_refined = self.D_F.train_on_batch(feature_refined, refined)
dfeature_loss = 0.5 * np.add(dfeature_loss_real, dfeature_loss_refined)
class_loss = self.Classifier.train_on_batch(feature_sim, classes)
gan1_loss = self.GAN_1.train_on_batch(imgs_sim, valid)
gan2_loss = self.GAN_2.train_on_batch(imgs_sim, valid)
gan3_loss = self.GAN_3.train_on_batch(imgs_sim, classes)
elapsed_time = datetime.datetime.now() - start_time
print ("[Epoch %d/%d] [targetatch %d/%d] [DR loss: %f] [DF loss: %f] [C loss: %f] [GAN_1 loss %f] [GAN_2 loss %f] [GAN_3 loss %f] time: %s " \
% ( epoch, epochs,
batch_i, self.data_loader.n_batches,
dimg_loss[0],
dfeature_loss[0],
class_loss[0],
gan1_loss[0],
gan2_loss[0],
gan3_loss[0],
elapsed_time))
if batch_i % interval == 0:
self.sample_images(epoch, batch_i)
def sample_images(self, epoch, batch_i):
os.makedirs('output/', exist_ok=True)
r, c = 1, 3
imgs_sim = self.data_loader.load_data(domain="sim", batch_size=1, is_testing=True)
imgs_target = self.data_loader.load_data(domain="target", batch_size=1, is_testing=True)
imgs_refined = self.Refiner.predict(imgs_sim)
gen_imgs = np.concatenate([imgs_sim, imgs_refined, imgs_target])
gen_imgs = 0.5 * gen_imgs + 0.5
titles = ['Simulated', 'Refined','Target']
fig, axs = plt.subplots(r, c)
axs[0].imshow(gen_imgs[0])
axs[0].set_title(titles[0])
axs[0].axis('off')
axs[1].imshow(gen_imgs[1])
axs[1].set_title(titles[1])
axs[1].axis('off')
axs[2].imshow(gen_imgs[2])
axs[2].set_title(titles[2])
axs[2].axis('off')
fig.savefig("output/%d_%d.png" % (epoch, batch_i))
plt.close()
if __name__ == '__main__':
cgan = CGAN()
cgan.train(epochs=100, batch_size=8, interval=50)