forked from FuxiCV/3D-Face-GCNs
-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_normal.py
180 lines (156 loc) · 6.72 KB
/
model_normal.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
import logging
import tensorflow as tf
from base_model import BaseModel
logger = logging.getLogger('x')
class Model(BaseModel):
"""
Mesh Convolutional Autoencoder which uses the Chebyshev approximation.
"""
def __init__(self, *args, **kwargs):
super(Model, self).__init__(*args, **kwargs)
logger.info('Using Normal Model...')
def mesh_generator(self, image_emb, pca_color, reuse=False):
with tf.variable_scope('mesh_generator', reuse=reuse):
decode_color = self.mesh_decoder(image_emb, reuse=reuse)
refine_color = self.mesh_refiner(pca_color, reuse=reuse)
with tf.variable_scope('mesh_concat'):
concat = tf.concat([decode_color, refine_color], axis=-1)
outputs = self.chebyshev5(concat, self.laplacians[0], 3, 6)
outputs = tf.nn.tanh(outputs)
return outputs
def mesh_decoder(self, image_emb, reuse=False):
if self.wide:
F = [32, 32, 64, 128, 256]
else:
F = [32, 16, 16, 16, 16]
with tf.variable_scope('mesh_decoder', reuse=reuse):
with tf.variable_scope('fc'):
x = self.fc(image_emb, self.pool_size[-1] * F[0]) # N x MF
x = tf.reshape(x,
[self.batch_size, self.pool_size[-1], F[0]]) # N x M x F
for i in range(4):
with tf.variable_scope('upconv{}'.format(i + 1)):
with tf.name_scope('unpooling'):
x = self.unpool(x, self.upsamp_trans[-i - 1])
with tf.name_scope('filter'):
x = self.chebyshev5(x, self.laplacians[-i - 2], F[i + 1], 6)
with tf.name_scope('bias_relu'):
x = self.brelu(x)
with tf.name_scope('outputs'):
x = self.chebyshev5(x, self.laplacians[0], 3, 6)
x = self.brelu(x)
# outputs = tf.nn.tanh(x)
return x
# def mesh_refiner(self, pca_color, reuse=False):
# if self.wide:
# F = [16, 32, 64, 128]
# else:
# F = [16, 32, 32, 16]
# with tf.variable_scope('mesh_refiner', reuse=reuse):
# x = pca_color
# for i in range(4):
# with tf.variable_scope('graph_conv{}'.format(i + 1)):
# with tf.name_scope('filter'):
# x = self.chebyshev5(x, self.laplacians[0], F[i], 6)
# with tf.name_scope('bias_relu'):
# x = self.brelu(x)
# with tf.name_scope('outputs'):
# x = self.chebyshev5(x, self.laplacians[0], 3, 6)
# x = self.brelu(x)
# # outputs = tf.nn.tanh(x)
# return x
# def mesh_refiner(self, pca_color, reuse=False):
# if self.wide:
# F = [16, 32, 64, 128]
# else:
# F = [16, 32, 64]
# with tf.variable_scope('mesh_refiner', reuse=reuse):
# x = pca_color
# with tf.variable_scope('graph_conv0'):
# with tf.name_scope('filter'):
# x = self.chebyshev5(x, self.laplacians[0], F[0], 6)
# for i in range(3):
# with tf.variable_scope('graph_conv{}'.format(i + 1)):
# with tf.name_scope('pooling'):
# x = self.unpool(x, self.downsamp_trans[i])
# with tf.name_scope('filter'):
# x = self.chebyshev5(x, self.laplacians[i + 1], F[i], 6)
# with tf.name_scope('bias_relu'):
# x = self.brelu(x)
# for i in range(3):
# with tf.variable_scope('graph_conv{}'.format(i + 4)):
# with tf.name_scope('unpooling'):
# x = self.unpool(x, self.upsamp_trans[-i - 2])
# with tf.name_scope('filter'):
# x = self.chebyshev5(x, self.laplacians[-i - 3], F[-i - 1], 6)
# with tf.name_scope('bias_relu'):
# x = self.brelu(x)
# with tf.name_scope('outputs'):
# x = self.chebyshev5(x, self.laplacians[0], 3, 6)
# x = self.brelu(x)
# # outputs = tf.nn.tanh(x)
# return x
def mesh_refiner(self, pca_color, reuse=False):
if self.wide:
F = [16, 32, 64, 128]
else:
F = [16, 32, 64, 128]
with tf.variable_scope('mesh_refiner', reuse=reuse):
x = pca_color
with tf.variable_scope('graph_conv0'):
with tf.name_scope('filter'):
x = self.chebyshev5(x, self.laplacians[0], F[0], 6)
layer_enc = []
for i in range(4):
with tf.variable_scope('graph_conv{}'.format(i + 1)):
with tf.name_scope('pooling'):
x = self.unpool(x, self.downsamp_trans[i])
with tf.name_scope('filter'):
x = self.chebyshev5(x, self.laplacians[i + 1], F[i], 6)
with tf.name_scope('bias_relu'):
x = self.brelu(x)
layer_enc.append(x)
x = tf.reshape(x, [self.batch_size, self.pool_size[-1] * F[-1]]) # N x MF
with tf.variable_scope('fc'):
x = self.fc(x, int(self.z_dim)) # N x M0
with tf.variable_scope('fc2'):
x = self.fc(x, self.pool_size[-1] * F[-1]) # N x MF
x = tf.reshape(x,
[self.batch_size, self.pool_size[-1], F[-1]]) # N x M x F
for i in range(4):
with tf.variable_scope('graph_conv{}'.format(i + 5)):
with tf.name_scope('unpooling'):
x = self.unpool(x, self.upsamp_trans[-i - 1])
#TODO: with skip or not
if i < 2:
x = tf.concat([x, layer_enc[-i - 2]], axis=-1)
with tf.name_scope('filter'):
x = self.chebyshev5(x, self.laplacians[-i - 2], F[-i - 1], 6)
with tf.name_scope('bias_relu'):
x = self.brelu(x)
with tf.name_scope('outputs'):
x = self.chebyshev5(x, self.laplacians[0], 3, 6)
x = self.brelu(x)
# outputs = tf.nn.tanh(x)
return x
def image_disc(self, inputs, t=True, reuse=False):
with tf.variable_scope('image_disc', reuse=reuse):
x = inputs
x = self.conv2d(x, 16, 1, 1, is_training=t, name='conv1_1')
# x = self.conv2d(x, 32, 3, 1, is_training=t, name='conv1_2')
x = tf.nn.max_pool(x, [1, 2, 2, 1], [1, 2, 2, 1], 'SAME')
x = self.conv2d(x, 32, 3, 1, is_training=t, name='conv2_1')
# x = self.conv2d(x, 64, 3, 1, is_training=t, name='conv2_2')
x = tf.nn.max_pool(x, [1, 2, 2, 1], [1, 2, 2, 1], 'SAME')
x = self.conv2d(x, 64, 3, 1, is_training=t, name='conv3_1')
# x = self.conv2d(x, 128, 3, 1, is_training=t, name='conv3_2')
x = tf.nn.max_pool(x, [1, 2, 2, 1], [1, 2, 2, 1], 'SAME')
x = self.conv2d(x, 128, 3, 1, is_training=t, name='conv4_1')
# x = self.conv2d(x, 256, 3, 1, is_training=t, name='conv4_2')
x = tf.nn.max_pool(x, [1, 2, 2, 1], [1, 2, 2, 1], 'SAME')
x = self.conv2d(x, 256, 3, 1, is_training=t, name='conv5_1')
# x = self.conv2d(x, 512, 3, 1, is_training=t, name='conv5_2')
x = tf.nn.max_pool(x, [1, 2, 2, 1], [1, 2, 2, 1], 'SAME')
x = self.conv2d(x, 512, 3, 1, is_training=t, name='conv6_1')
x = self.conv2d(x, 1, 7, 1, 'VALID', False, False, t, 'outputs')
return tf.squeeze(x, axis=[1, 2])