/
model.py
179 lines (114 loc) · 7.16 KB
/
model.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
#################################
# RAW-to-RGB Model architecture #
#################################
import tensorflow as tf
import numpy as np
def PUNET(input, instance_norm=False, instance_norm_level_1=False, num_maps_base=16):
with tf.compat.v1.variable_scope("generator"):
# -----------------------------------------
# Downsampling layers
conv_l1_d1 = _conv_multi_block(input, 3, num_maps=num_maps_base, instance_norm=False) # 128 -> 128
pool1 = max_pool(conv_l1_d1, 2) # 128 -> 64
conv_l2_d1 = _conv_multi_block(pool1, 3, num_maps=num_maps_base*2, instance_norm=instance_norm) # 64 -> 64
pool2 = max_pool(conv_l2_d1, 2) # 64 -> 32
conv_l3_d1 = _conv_multi_block(pool2, 3, num_maps=num_maps_base*4, instance_norm=instance_norm) # 32 -> 32
pool3 = max_pool(conv_l3_d1, 2) # 32 -> 16
conv_l4_d1 = _conv_multi_block(pool3, 3, num_maps=num_maps_base*8, instance_norm=instance_norm) # 16 -> 16
pool4 = max_pool(conv_l4_d1, 2) # 16 -> 8
# -----------------------------------------
# Processing: Level 5, Input size: 8 x 8
conv_l5_d1 = _conv_multi_block(pool4, 3, num_maps=num_maps_base*16, instance_norm=instance_norm)
conv_l5_d2 = _conv_multi_block(conv_l5_d1, 3, num_maps=num_maps_base*16, instance_norm=instance_norm) + conv_l5_d1
conv_l5_d3 = _conv_multi_block(conv_l5_d2, 3, num_maps=num_maps_base*16, instance_norm=instance_norm) + conv_l5_d2
conv_l5_d4 = _conv_multi_block(conv_l5_d3, 3, num_maps=num_maps_base*16, instance_norm=instance_norm)
conv_t4b = _conv_tranpose_layer(conv_l5_d4, num_maps_base*8, 3, 2) # 8 -> 16
# -----------------------------------------
# Processing: Level 4, Input size: 16 x 16
conv_l4_d6 = conv_l4_d1
conv_l4_d7 = stack(conv_l4_d6, conv_t4b)
conv_l4_d8 = _conv_multi_block(conv_l4_d7, 3, num_maps=num_maps_base*8, instance_norm=instance_norm)
conv_t3b = _conv_tranpose_layer(conv_l4_d8, num_maps_base*4, 3, 2) # 16 -> 32
# -----------------------------------------
# Processing: Level 3, Input size: 32 x 32
conv_l3_d6 = conv_l3_d1
conv_l3_d7 = stack(conv_l3_d6, conv_t3b)
conv_l3_d8 = _conv_multi_block(conv_l3_d7, 3, num_maps=num_maps_base*4, instance_norm=instance_norm)
conv_t2b = _conv_tranpose_layer(conv_l3_d8, num_maps_base*2, 3, 2) # 32 -> 64
# -------------------------------------------
# Processing: Level 2, Input size: 64 x 64
conv_l2_d7 = conv_l2_d1
conv_l2_d8 = stack(_conv_multi_block(conv_l2_d7, 3, num_maps=num_maps_base*2, instance_norm=instance_norm), conv_t2b)
conv_l2_d9 = _conv_multi_block(conv_l2_d8, 3, num_maps=num_maps_base*2, instance_norm=instance_norm)
conv_t1b = _conv_tranpose_layer(conv_l2_d9, num_maps_base, 3, 2) # 64 -> 128
# -------------------------------------------
# Processing: Level 1, Input size: 128 x 128
conv_l1_d9 = conv_l1_d1
conv_l1_d10 = stack(_conv_multi_block(conv_l1_d9, 3, num_maps=num_maps_base, instance_norm=False), conv_t1b)
conv_l1_d11 = stack(conv_l1_d10, conv_l1_d1)
conv_l1_d12 = _conv_multi_block(conv_l1_d11, 3, num_maps=num_maps_base, instance_norm=False)
# ----------------------------------------------------------
# Processing: Level 0 (x2 upscaling), Input size: 128 x 128
conv_l0 = _conv_tranpose_layer(conv_l1_d12, num_maps_base//4, 3, 2) # 128 -> 256
conv_l0_out = _conv_layer(conv_l0, 3, 3, 1, relu=False, instance_norm=False)
output_l0 = tf.nn.tanh(conv_l0_out) * 0.58 + 0.5
output_l0 = tf.identity(output_l0, name='output_l0')
return output_l0
def _conv_multi_block(input, max_size, num_maps, instance_norm):
conv_3a = _conv_layer(input, num_maps, 3, 1, relu=True, instance_norm=instance_norm)
conv_3b = _conv_layer(conv_3a, num_maps, 3, 1, relu=True, instance_norm=instance_norm)
output_tensor = conv_3b
if max_size >= 5:
conv_5a = _conv_layer(input, num_maps, 5, 1, relu=True, instance_norm=instance_norm)
conv_5b = _conv_layer(conv_5a, num_maps, 5, 1, relu=True, instance_norm=instance_norm)
output_tensor = stack(output_tensor, conv_5b)
if max_size >= 7:
conv_7a = _conv_layer(input, num_maps, 7, 1, relu=True, instance_norm=instance_norm)
conv_7b = _conv_layer(conv_7a, num_maps, 7, 1, relu=True, instance_norm=instance_norm)
output_tensor = stack(output_tensor, conv_7b)
if max_size >= 9:
conv_9a = _conv_layer(input, num_maps, 9, 1, relu=True, instance_norm=instance_norm)
conv_9b = _conv_layer(conv_9a, num_maps, 9, 1, relu=True, instance_norm=instance_norm)
output_tensor = stack(output_tensor, conv_9b)
return output_tensor
def stack(x, y):
return tf.concat([x, y], 3)
def conv2d(x, W):
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
def _conv_layer(net, num_filters, filter_size, strides, relu=True, instance_norm=False, padding='SAME'):
weights_init = _conv_init_vars(net, num_filters, filter_size)
strides_shape = [1, strides, strides, 1]
bias = tf.Variable(tf.constant(0.01, shape=[num_filters]))
net = tf.nn.conv2d(net, weights_init, strides_shape, padding=padding) + bias
if instance_norm:
net = _instance_norm(net)
if relu:
net = tf.compat.v1.nn.leaky_relu(net)
return net
def _instance_norm(net):
batch, rows, cols, channels = [i.value for i in net.get_shape()]
var_shape = [channels]
mu, sigma_sq = tf.nn.moments(net, [1,2], keep_dims=True)
shift = tf.Variable(tf.zeros(var_shape))
scale = tf.Variable(tf.ones(var_shape))
epsilon = 1e-3
normalized = (net-mu)/(sigma_sq + epsilon)**(.5)
return scale * normalized + shift
def _conv_init_vars(net, out_channels, filter_size, transpose=False):
_, rows, cols, in_channels = [i.value for i in net.get_shape()]
if not transpose:
weights_shape = [filter_size, filter_size, in_channels, out_channels]
else:
weights_shape = [filter_size, filter_size, out_channels, in_channels]
weights_init = tf.Variable(tf.random.truncated_normal(weights_shape, stddev=0.01, seed=1), dtype=tf.float32)
return weights_init
def _conv_tranpose_layer(net, num_filters, filter_size, strides):
weights_init = _conv_init_vars(net, num_filters, filter_size, transpose=True)
batch_size, rows, cols, in_channels = [i.value for i in net.get_shape()]
new_rows, new_cols = int(rows * strides), int(cols * strides)
new_shape = [batch_size, new_rows, new_cols, num_filters]
tf_shape = tf.stack(new_shape)
strides_shape = [1, strides, strides, 1]
net = tf.nn.conv2d_transpose(net, weights_init, tf_shape, strides_shape, padding='SAME')
return tf.compat.v1.nn.leaky_relu(net)
def max_pool(x, n):
return tf.nn.max_pool(x, ksize=[1, n, n, 1], strides=[1, n, n, 1], padding='VALID')