-
Notifications
You must be signed in to change notification settings - Fork 0
/
CUB128GANAE.py
268 lines (238 loc) · 11 KB
/
CUB128GANAE.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
import tensorflow as tf
from layers import conv2d, linear, flatten, nnupsampling, batchnorm, gaussnoise, pool
from activations import lrelu
from op import log_sum_exp
from data_loader import train_loader, validation_loader
from neon.backends import gen_backend
import numpy as np
from utils import drawblock, createfolders, OneHot, image_reshape
from scipy.misc import imsave
import os
# Create folders to store images
gen_dir, real_dir, gen_dir128 = createfolders("./genimgs/CUB128GANAE", "/gen", "/real", "/gen128")
# Create folder to store models
dir_name = './models/CUB128GANAE'
if not os.path.exists(dir_name):
os.mkdir(dir_name)
# Parameters
init_iter, max_iter = 0, 30000
display_iter = 100
eval_iter = 100
store_img_iter = 100
save_iter = 1000
lr_init = 0.0002
batch_size = 100
zdim = 100
n_classes = 200
dropout = 0.2
im_size = [64, 64]
dname, gname = 'd_', 'g_'
tf.set_random_seed(1234)
# DataLoader
be = gen_backend(backend='cpu', batch_size=batch_size, datatype=np.float32)
root_files = './dataset/cub200'
manifestfile = os.path.join(root_files, 'train-index.csv')
testmanifest = os.path.join(root_files, 'val-index.csv')
train = train_loader(manifestfile, root_files, be, h=im_size[0], w=im_size[1], scale=[0.875, 0.875])
test = validation_loader(testmanifest, root_files, be, h=im_size[0], w=im_size[1], scale=[0.875, 0.875], ncls=n_classes)
OneHot = OneHot(be, n_classes)
# Graph input
is_train = tf.placeholder(tf.bool)
keep_prob = tf.placeholder(tf.float32)
x_n = tf.placeholder(tf.float32, [batch_size, 3, im_size[0], im_size[1]])
y = tf.placeholder(tf.float32, [batch_size, n_classes])
lr_tf = tf.placeholder(tf.float32)
z = tf.random_uniform([batch_size, zdim], -1, 1)
iny = tf.placeholder(tf.float32, [batch_size, n_classes])
# Discriminator
def discriminator(inp, reuse=False):
with tf.variable_scope('Encoder', reuse=reuse):
# 64
inp = gaussnoise(inp, std=0.05)
conv1 = conv2d(inp, 64, kernel=3, strides=2, name=dname + 'conv1')
conv1 = lrelu(conv1, 0.2)
# 32
conv2 = tf.nn.dropout(conv1, keep_prob)
conv2 = conv2d(conv2, 128, kernel=3, strides=2, name=dname + 'conv2')
conv2 = batchnorm(conv2, is_training=is_train, name=dname + 'bn2')
conv2 = lrelu(conv2, 0.2)
# 16
conv3 = tf.nn.dropout(conv2, keep_prob)
conv3 = conv2d(conv3, 256, kernel=3, strides=2, name=dname + 'conv3')
conv3 = batchnorm(conv3, is_training=is_train, name=dname + 'bn3')
conv3 = lrelu(conv3, 0.2)
# 8
conv3b = conv2d(conv3, 256, kernel=3, strides=1, name=dname + 'conv3b')
conv3b = batchnorm(conv3b, is_training=is_train, name=dname + 'bn3b')
conv3b = lrelu(conv3b, 0.2)
conv4 = tf.nn.dropout(conv3b, keep_prob)
conv4 = conv2d(conv4, 512, kernel=3, strides=2, name=dname + 'conv4')
conv4 = batchnorm(conv4, is_training=is_train, name=dname + 'bn4')
conv4 = lrelu(conv4, 0.2)
# 4
conv4b = conv2d(conv4, 512, kernel=3, strides=1, name=dname + 'conv4b')
conv4b = batchnorm(conv4b, is_training=is_train, name=dname + 'bn4b')
conv4b = lrelu(conv4b, 0.2)
flat = flatten(conv4b)
# Classifier
clspred = linear(flat, n_classes, name=dname + 'cpred')
# Decoder
g1 = conv2d(conv4b, nout=512, kernel=3, name=dname + 'deconv1')
g1 = batchnorm(g1, is_training=tf.constant(True), name=dname + 'bn1g')
g1 = lrelu(g1, 0.2)
g2 = nnupsampling(g1, [8, 8])
g2 = conv2d(g2, nout=256, kernel=3, name=dname + 'deconv2')
g2 = batchnorm(g2, is_training=tf.constant(True), name=dname + 'bn2g')
g2 = lrelu(g2, 0.2)
g3 = nnupsampling(g2, [16, 16])
g3 = conv2d(g3, nout=128, kernel=3, name=dname + 'deconv3')
g3 = batchnorm(g3, is_training=tf.constant(True), name=dname + 'bn3g')
g3 = lrelu(g3, 0.2)
g4 = nnupsampling(g3, [32, 32])
g4 = conv2d(g4, nout=64, kernel=3, name=dname + 'deconv4')
g4 = batchnorm(g4, is_training=tf.constant(True), name=dname + 'bn4g')
g4 = lrelu(g4, 0.2)
g5 = nnupsampling(g4, [64, 64])
g5 = conv2d(g5, nout=32, kernel=3, name=dname + 'deconv5')
g5 = batchnorm(g5, is_training=tf.constant(True), name=dname + 'bn5g')
g5 = lrelu(g5, 0.2)
g5b = conv2d(g5, nout=3, kernel=3, name=dname + 'deconv5b')
g5b = tf.nn.tanh(g5b)
return clspred, g5b, flat
# Generator
def generator(inp_z, inp_y, reuse=False):
with tf.variable_scope('Generator', reuse=reuse):
inp = tf.concat([inp_z, inp_y], 1)
sz = 4
g1 = linear(inp, 512 * sz * sz, name=gname + 'deconv1')
g1 = batchnorm(g1, is_training=tf.constant(True), name=gname + 'bn1g')
g1 = lrelu(g1, 0.2)
g1_reshaped = tf.reshape(g1, [-1, 512, sz, sz])
print 'genreshape: ' + str(g1_reshaped.get_shape().as_list())
g2 = nnupsampling(g1_reshaped, [8, 8])
g2 = conv2d(g2, nout=512, kernel=3, name=gname + 'deconv2')
g2 = batchnorm(g2, is_training=tf.constant(True), name=gname + 'bn2g')
g2 = lrelu(g2, 0.2)
g3 = nnupsampling(g2, [16, 16])
g3 = conv2d(g3, nout=256, kernel=3, name=gname + 'deconv3')
g3 = batchnorm(g3, is_training=tf.constant(True), name=gname + 'bn3g')
g3 = lrelu(g3, 0.2)
g4 = nnupsampling(g3, [32, 32])
g4 = conv2d(g4, nout=128, kernel=3, name=gname + 'deconv4')
g4 = batchnorm(g4, is_training=tf.constant(True), name=gname + 'bn4g')
g4 = lrelu(g4, 0.2)
g5 = nnupsampling(g4, [64, 64])
g5 = conv2d(g5, nout=64, kernel=3, name=gname + 'deconv5')
g5 = batchnorm(g5, is_training=tf.constant(True), name=gname + 'bn5g')
g5 = lrelu(g5, 0.2)
g5b = conv2d(g5, nout=64, kernel=3, name=gname + 'deconv5b')
g5b = batchnorm(g5b, is_training=tf.constant(True), name=gname + 'bn5bg')
g5b = lrelu(g5b, 0.2)
g6 = nnupsampling(g5b, [128, 128])
g6 = conv2d(g6, nout=32, kernel=3, name=gname + 'deconv6')
g6 = batchnorm(g6, is_training=tf.constant(True), name=gname + 'bn6g')
g6 = lrelu(g6, 0.2)
g6b = conv2d(g6, nout=3, kernel=3, name=gname + 'deconv6b')
g6b = tf.nn.tanh(g6b)
g6b_64 = pool(g6b, fsize=3, strides=2, op='avg')
return g6b_64, g6b
# Call functions
Opred_n, recon_n, _ = discriminator(x_n)
samples, samples128 = generator(z, iny)
Opred_g, recon_g, embed = discriminator(samples, reuse=True)
# Get trainable variables and split
t_vars = tf.trainable_variables()
d_vars = [var for var in t_vars if dname in var.name]
g_vars = [var for var in t_vars if gname in var.name]
print [var.name for var in d_vars]
print [var.name for var in g_vars]
# Define D loss
lreal = log_sum_exp(Opred_n)
lfake = log_sum_exp(Opred_g)
cost_On = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=Opred_n, labels=y))
cost_Dn = - tf.reduce_mean(lreal) + tf.reduce_mean(tf.nn.softplus(lreal))
cost_Dg_fake = tf.reduce_mean(tf.nn.softplus(lfake))
cost_msen = tf.reduce_mean(tf.square(recon_n - x_n)) * 0.5
cost_mseg = tf.reduce_mean(tf.square(recon_g - samples)) * 0.5
D_loss = cost_On + cost_Dn + cost_Dg_fake + cost_msen
# Define G loss
cost_Dg = - tf.reduce_mean(lfake) + tf.reduce_mean(tf.nn.softplus(lfake))
cost_Og = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=Opred_g, labels=iny))
G_loss = cost_Dg + cost_Og + cost_mseg
# Define optimizer
d_optimizer = tf.train.AdamOptimizer(learning_rate=lr_tf, beta1=0.5).minimize(D_loss, var_list=d_vars)
g_optimizer = tf.train.AdamOptimizer(learning_rate=lr_tf, beta1=0.5).minimize(G_loss, var_list=g_vars)
# Evaluate model
Oaccuracy = tf.reduce_mean(tf.cast(tf.equal(tf.argmax(Opred_n, 1), tf.argmax(y, 1)), tf.float32))
# Initialize the variables
init = tf.global_variables_initializer()
# Reset train dataset
train.reset()
# Config for session
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
# Train
with tf.Session(config=config) as sess:
sess.run(init)
saver = tf.train.Saver(max_to_keep=None)
coord = tf.train.Coordinator()
threads = tf.train.start_queue_runners(coord=coord)
for i_iter in range(init_iter, max_iter):
# Control lr
if i_iter < 15000:
lr = lr_init
else:
lr = lr_init / 10.
# Set current fake label
inds = np.repeat(np.random.permutation(n_classes)[:20], 5)
fake_y = np.zeros((batch_size, n_classes))
fake_y[np.arange(batch_size), inds] = 1.
# Fetch minibatch
batch_x, batch_y = train.next()
batch_x = image_reshape(batch_x.get(), im_size, input_format='tanh')
batch_y = OneHot.transform(batch_y).get().transpose()
# update discriminator
_, lossDn, lossOn, lossFake = sess.run([d_optimizer, cost_Dn, cost_On, cost_Dg_fake], feed_dict={
x_n: batch_x, y: batch_y, iny: fake_y,
keep_prob: 1. - dropout, is_train: True, lr_tf: lr
})
# update generator
_, gen_img, gen_img128 = sess.run([g_optimizer, samples, samples128], feed_dict={
iny: fake_y,
keep_prob: 1., is_train: True, lr_tf: lr
})
# print losses
if i_iter % display_iter == 0 or i_iter == max_iter - 1:
print 'Iteration: %i, lossDn: %.2f, lossOn: %.2f, lossFake: %.2f' % (i_iter, lossDn, lossOn, lossFake)
# Evaluate classification accuracy
if i_iter % eval_iter == 0 or i_iter == max_iter - 1:
total_Oaccuracy = 0.
test.reset()
for mb_idx, (batch_x, batch_y) in enumerate(test):
batch_x = image_reshape(batch_x.get(), im_size, input_format='tanh')
batch_y = batch_y.get().transpose()
total_Oaccuracy += sess.run(Oaccuracy,
feed_dict={x_n: batch_x, y: batch_y, keep_prob: 1., is_train: False})
print 'Iteration %i, Accuracy: %.2f' % (i_iter, total_Oaccuracy / mb_idx)
# Store images
if i_iter % store_img_iter == 0 or i_iter == max_iter - 1:
# Store Generated
genmix_imgs = (np.transpose(gen_img, [0, 2, 3, 1]) + 1.) * 127.5
genmix_imgs = np.uint8(genmix_imgs[:, :, :, ::-1])
genmix_imgs = drawblock(genmix_imgs, 10)
imsave(os.path.join(gen_dir, '%i.jpg' % i_iter), genmix_imgs)
# Store Generated 96
genmix_imgs = (np.transpose(gen_img128, [0, 2, 3, 1]) + 1.) * 127.5
genmix_imgs = np.uint8(genmix_imgs[:, :, :, ::-1])
genmix_imgs = drawblock(genmix_imgs, 10)
imsave(os.path.join(gen_dir128, '%i.jpg' % i_iter), genmix_imgs)
# Store Real
real_imgs = (np.transpose(batch_x, [0, 2, 3, 1]) + 1.) * 127.5
real_imgs = np.uint8(real_imgs[:, :, :, ::-1])
real_imgs = drawblock(real_imgs, 10)
imsave(os.path.join(real_dir, '%i.jpg' % i_iter), real_imgs)
# Store model
if i_iter % save_iter == 0 or i_iter == max_iter - 1 or i_iter == max_iter:
save_path = saver.save(sess, dir_name + '/cdgan%i.ckpt' % i_iter)
coord.request_stop()
coord.join(threads)