-
Notifications
You must be signed in to change notification settings - Fork 37
/
Copy pathmain_3D-RecGAN.py
218 lines (186 loc) · 10.3 KB
/
main_3D-RecGAN.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
import os
import shutil
import tensorflow as tf
import scipy.io
import tools
resolution = 64
batch_size = 8
GPU0 = '1'
###############################################################
config={}
config['train_names'] = ['chair']
for name in config['train_names']:
config['X_train_'+name] = './Data/'+name+'/train_25d/voxel_grids_64/'
config['Y_train_'+name] = './Data/'+name+'/train_3d/voxel_grids_64/'
config['test_names']=['chair']
for name in config['test_names']:
config['X_test_'+name] = './Data/'+name+'/test_25d/voxel_grids_64/'
config['Y_test_'+name] = './Data/'+name+'/test_3d/voxel_grids_64/'
config['resolution'] = resolution
config['batch_size'] = batch_size
################################################################
class Network:
def __init__(self):
self.train_mod_dir = './train_mod/'
self.train_sum_dir = './train_sum/'
self.test_res_dir = './test_res/'
self.test_sum_dir = './test_sum/'
if os.path.exists(self.test_res_dir):
shutil.rmtree(self.test_res_dir)
print ('test_res_dir: deleted and then created!')
os.makedirs(self.test_res_dir)
if os.path.exists(self.train_mod_dir):
shutil.rmtree(self.train_mod_dir)
print ('train_mod_dir: deleted and then created!')
os.makedirs(self.train_mod_dir)
if os.path.exists(self.train_sum_dir):
shutil.rmtree(self.train_sum_dir)
print ('train_sum_dir: deleted and then created!')
os.makedirs(self.train_sum_dir)
if os.path.exists(self.test_sum_dir):
shutil.rmtree(self.test_sum_dir)
print ('test_sum_dir: deleted and then created!')
os.makedirs(self.test_sum_dir)
def ae_u(self,X):
with tf.device('/gpu:'+GPU0):
X = tf.reshape(X,[batch_size,resolution,resolution,resolution,1])
##### encode
c_e = [1, 64, 128, 256, 512]
s_e = [0, 1, 1, 1, 1]
layers_e = []
layers_e.append(X)
for i in range(1, 5, 1):
layer = tools.Ops.conv3d(layers_e[-1], k=4, out_c=c_e[i], str=s_e[i], name='e' + str(i))
layer = tools.Ops.maxpool3d(tools.Ops.xxlu(layer,name='lrelu'), k=2, s=2, pad='SAME')
layers_e.append(layer)
##### fc
bat, d1, d2, d3, cc = [int(d) for d in layers_e[-1].get_shape()]
lfc = tf.reshape(layers_e[-1], [bat, -1])
lfc = tools.Ops.xxlu(tools.Ops.fc(lfc, out_d=5000, name='fc1'), name='relu')
with tf.device('/gpu:'+GPU0):
lfc = tools.Ops.xxlu(tools.Ops.fc(lfc, out_d=d1 * d2 * d3 * cc, name='fc2'),name='relu')
lfc = tf.reshape(lfc, [bat, d1, d2, d3, cc])
##### decode
c_d = [0, 256, 128, 64, 1]
s_d = [0, 2, 2, 2, 2, 2]
layers_d = []
layers_d.append(lfc)
for j in range(1, 5, 1):
u_net = True
if u_net:
layer = tf.concat([layers_d[-1], layers_e[-j]], axis=4)
layer = tools.Ops.deconv3d(layer, k=4, out_c=c_d[j], str=s_d[j], name='d' + str(len(layers_d)))
else:
layer = tools.Ops.deconv3d(layers_d[-1], k=4, out_c=c_d[j], str=s_d[j], name='d' + str(len(layers_d)))
if j != 4:
layer = tools.Ops.xxlu(layer,name='relu')
layers_d.append(layer)
vox_sig = tf.sigmoid(layers_d[-1])
vox_sig_modified = tf.maximum(vox_sig,0.01)
return vox_sig, vox_sig_modified
def dis(self, X, Y):
with tf.device('/gpu:'+GPU0):
X = tf.reshape(X,[batch_size,resolution,resolution,resolution,1])
Y = tf.reshape(Y,[batch_size,resolution,resolution,resolution,1])
layer = tf.concat([X,Y],axis=4)
c_d = [1,64,128,256,512]
s_d = [0,2,2,2,2]
layers_d =[]
layers_d.append(layer)
for i in range(1,5,1):
layer = tools.Ops.conv3d(layers_d[-1],k=4,out_c=c_d[i],str=s_d[i],name='d'+str(i))
if i!=4:
layer = tools.Ops.xxlu(layer, name='lrelu')
layers_d.append(layer)
y = tf.reshape(layers_d[-1],[batch_size,-1])
return tf.nn.sigmoid(y)
def train(self, data):
X = tf.placeholder(shape=[batch_size, resolution, resolution, resolution, 1], dtype=tf.float32)
Y = tf.placeholder(shape=[batch_size, resolution, resolution, resolution, 1], dtype=tf.float32)
lr = tf.placeholder(tf.float32)
with tf.variable_scope('ae'):
Y_pred, Y_pred_modi = self.ae_u(X)
with tf.variable_scope('dis'):
XY_real_pair = self.dis(X, Y)
with tf.variable_scope('dis',reuse=True):
XY_fake_pair = self.dis(X, Y_pred)
with tf.device('/gpu:'+GPU0):
################################ ae loss
Y_ = tf.reshape(Y,shape=[batch_size,-1])
Y_pred_modi_ = tf.reshape(Y_pred_modi,shape=[batch_size,-1])
w = 0.85
ae_loss = tf.reduce_mean( -tf.reduce_mean(w*Y_*tf.log(Y_pred_modi_ + 1e-8),reduction_indices=[1]) -
tf.reduce_mean((1-w)*(1-Y_)*tf.log(1-Y_pred_modi_ + 1e-8), reduction_indices=[1]) )
sum_ae_loss = tf.summary.scalar('ae_loss', ae_loss)
################################ wgan loss
gan_g_loss = -tf.reduce_mean(XY_fake_pair)
gan_d_loss_no_gp = tf.reduce_mean(XY_fake_pair) - tf.reduce_mean(XY_real_pair)
sum_gan_g_loss = tf.summary.scalar('gan_g_loss',gan_g_loss)
sum_gan_d_loss_no_gp = tf.summary.scalar('gan_d_loss_no_gp',gan_d_loss_no_gp)
alpha = tf.random_uniform(shape=[batch_size,resolution**3],minval=0.0,maxval=1.0)
Y_pred_ = tf.reshape(Y_pred,shape=[batch_size,-1])
differences_ = Y_pred_ -Y_
interpolates = Y_ + alpha*differences_
with tf.variable_scope('dis',reuse=True):
XY_fake_intep = self.dis(X, interpolates)
gradients = tf.gradients(XY_fake_intep,[interpolates])[0]
slopes = tf.sqrt(tf.reduce_sum(tf.square(gradients),reduction_indices=[1]))
gradient_penalty = tf.reduce_mean((slopes-1.0)**2)
sum_gp = tf.summary.scalar('wgan_gp', gradient_penalty)
gan_d_loss_gp = gan_d_loss_no_gp+10*gradient_penalty
################################# ae + gan loss
gan_g_w = 5
ae_w = 100-gan_g_w
ae_gan_g_loss = ae_w * ae_loss + gan_g_w * gan_g_loss
with tf.device('/gpu:' + GPU0):
ae_var = [var for var in tf.trainable_variables() if var.name.startswith('ae')]
dis_var = [var for var in tf.trainable_variables() if var.name.startswith('dis')]
ae_g_optim = tf.train.AdamOptimizer(learning_rate=lr, beta1=0.9, beta2=0.999, epsilon=1e-8).minimize(ae_gan_g_loss, var_list=ae_var)
dis_optim = tf.train.AdamOptimizer(learning_rate=lr,beta1=0.9,beta2=0.999,epsilon=1e-8).minimize(gan_d_loss_gp,var_list=dis_var)
print tools.Ops.variable_count()
sum_merged = tf.summary.merge_all()
saver = tf.train.Saver(max_to_keep=1)
config = tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.visible_device_list = GPU0
with tf.Session(config=config) as sess:
sum_writer_train = tf.summary.FileWriter(self.train_sum_dir, sess.graph)
sum_write_test = tf.summary.FileWriter(self.test_sum_dir)
if os.path.isfile(self.train_mod_dir + 'model.cptk.data-00000-of-00001'):
print ('restoring saved model')
saver.restore(sess, self.train_mod_dir + 'model.cptk')
else:
sess.run(tf.global_variables_initializer())
for epoch in range(15):
data.shuffle_X_Y_files(label='train')
total_train_batch_num = data.total_train_batch_num
print ('total_train_batch_num:', total_train_batch_num)
for i in range(total_train_batch_num):
#### training
X_train_batch, Y_train_batch = data.load_X_Y_voxel_grids_train_next_batch()
sess.run([dis_optim], feed_dict={X: X_train_batch, Y: Y_train_batch, lr: 0.0001})
sess.run([ae_g_optim], feed_dict={X: X_train_batch, Y: Y_train_batch, lr: 0.0005})
ae_loss_c,gan_g_loss_c,gan_d_loss_no_gp_c,gan_d_loss_gp_c,sum_train = sess.run(
[ae_loss, gan_g_loss, gan_d_loss_no_gp, gan_d_loss_gp,sum_merged],feed_dict={X: X_train_batch, Y: Y_train_batch})
if i%100==0:
sum_writer_train.add_summary(sum_train, epoch * total_train_batch_num + i)
print ('epoch:', epoch, 'i:', i, 'train ae loss:', ae_loss_c,'gan g loss:', gan_g_loss_c,
'gan d loss no gp:',gan_d_loss_no_gp_c, 'gan d loss gp:', gan_d_loss_gp_c)
#### testing
if i % 300 == 0 and epoch % 1 == 0:
X_test_batch, Y_test_batch = data.load_X_Y_voxel_grids_test_next_batch(fix_sample=False)
ae_loss_t,gan_g_loss_t,gan_d_loss_no_gp_t,gan_d_loss_gp_t, Y_test_pred = sess.run(
[ae_loss, gan_g_loss, gan_d_loss_no_gp,gan_d_loss_gp, Y_pred],feed_dict={X: X_test_batch, Y: Y_test_batch})
to_save = {'X_test': X_test_batch, 'Y_test_pred': Y_test_pred,'Y_test_true': Y_test_batch}
scipy.io.savemat(self.test_res_dir + 'X_Y_pred_' + str(epoch).zfill(2) + '_' + str(i).zfill(4) + '.mat', to_save, do_compression=True)
print ('epoch:', epoch, 'i:', i, 'test ae loss:', ae_loss_t, 'gan g loss:',gan_g_loss_t,
'gan d loss no gp:', gan_d_loss_no_gp_t, 'gan d loss gp:', gan_d_loss_gp_t)
#### full testing
# ...
#### model saving
if i %500==0 and i>0 and epoch % 1 == 0:
saver.save(sess, save_path=self.train_mod_dir +'model.cptk')
print "epoch:", epoch, " i:", i, " model saved!"
if __name__ == "__main__":
data = tools.Data(config)
net = Network()
net.train(data)