-
Notifications
You must be signed in to change notification settings - Fork 18
/
model.py
137 lines (113 loc) · 7.94 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
import tensorflow as tf
import tflearn
from tflearn.layers.core import input_data, dropout, fully_connected
from tflearn.layers.estimator import regression
from tflearn.layers.normalization import l2_normalize
from tflearn.metrics import Top_k, R2
def GenerateNet():
LeakyReluSlope = 0.3
#Defining the network
# input image 3 multiscale layers
# Current size: 256x256x3
NetworkIN = input_data(shape=[None, 256, 256, 3], name='input')
# Current size: 256x256x3
Network128 = tflearn.layers.conv.conv_2d(NetworkIN, 6, 4, strides=2, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network128)
# Current size: 128x128x6
Network64 = tflearn.layers.conv.conv_2d(Network128, 12, 4, strides=2, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network64)
# Current size: 64x64x12
Network32 = tflearn.layers.conv.conv_2d(Network64, 24, 4, strides=2, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network32)
# Current size: 32x32x24
Network16 = tflearn.layers.conv.conv_2d(Network32, 48, 4, strides=2, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network16)
# Current size: 16x16x48
Network8 = tflearn.layers.conv.conv_2d(Network16, 96, 4, strides=2, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network8)
# Current size: 8x8x96
Network4 = tflearn.layers.conv.conv_2d(Network8, 192, 4, strides=2, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network4)
# Current size: 4x4x192
Network2 = tflearn.layers.conv.conv_2d(Network4, 384, 4, strides=2, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network2)
#===================== UPSAMPLING DECODER ==========================
#current size: 2x2x384
#Network = tflearn.layers.core.dropout(Network2, 0.5)
#------------ 2x2 -> 4x4 ----------------------------------------------------------|
#current size: 2x2x384
Network = tflearn.layers.conv.upsample_2d(Network2, 2)
Network = tflearn.layers.conv.conv_2d(Network, 192, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#current size: 4x4x192
Network = tflearn.layers.merge_ops.merge((Network4, Network), mode='concat', axis=3)
#current size: 4x4x384
Network = tflearn.layers.conv.conv_2d(Network, 192, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#Network = tflearn.layers.core.dropout(Network, 0.5)
#------------ 4x4 -> 8x8 ----------------------------------------------------------|
#current size: 4x4x192
Network = tflearn.layers.conv.upsample_2d(Network, 2)
Network = tflearn.layers.conv.conv_2d(Network, 96, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#current size: 8x8x96
Network = tflearn.layers.merge_ops.merge((Network8, Network), mode='concat', axis=3)
#current size: 8x8x192
Network = tflearn.layers.conv.conv_2d(Network, 96, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#Network = tflearn.layers.core.dropout(Network, 0.5)
#------------ 8x8 -> 16x16 ----------------------------------------------------------|
#current size: 8x8x96
Network = tflearn.layers.conv.upsample_2d(Network, 2)
Network = tflearn.layers.conv.conv_2d(Network, 48, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#current size: 16x16x48
Network = tflearn.layers.merge_ops.merge((Network16, Network), mode='concat', axis=3)
#current size: 16x16x96
Network = tflearn.layers.conv.conv_2d(Network, 48, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#------------ 16x16 -> 32x32 ----------------------------------------------------------|
#current size: 16x16x48
Network = tflearn.layers.conv.upsample_2d(Network, 2)
Network = tflearn.layers.conv.conv_2d(Network, 24, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#current size: 32x32x24
Network = tflearn.layers.merge_ops.merge((Network32, Network), mode='concat', axis=3)
#current size: 32x32x48
Network = tflearn.layers.conv.conv_2d(Network, 24, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#------------ 32x32 -> 64x64 ----------------------------------------------------------|
#current size: 32x32x24
Network = tflearn.layers.conv.upsample_2d(Network, 2)
Network = tflearn.layers.conv.conv_2d(Network, 12, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#current size: 64x64x12
Network = tflearn.layers.merge_ops.merge((Network64, Network), mode='concat', axis=3)
#current size: 64x64x24
Network = tflearn.layers.conv.conv_2d(Network, 12, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#------------ 64x64 -> 128x128 ----------------------------------------------------------|
#current size: 64x64x12
Network = tflearn.layers.conv.upsample_2d(Network, 2)
Network = tflearn.layers.conv.conv_2d(Network, 6, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#current size: 128x128x6
Network = tflearn.layers.merge_ops.merge((Network128, Network), mode='concat', axis=3)
#current size: 128x128x12
Network = tflearn.layers.conv.conv_2d(Network, 6, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#------------ 128x128 -> 256x256 ----------------------------------------------------------|
#current size: 128x128x6
Network = tflearn.layers.conv.upsample_2d(Network, 2)
Network = tflearn.layers.conv.conv_2d(Network, 3, 4, strides=1, activation=lambda x: tflearn.activations.leaky_relu(x, alpha=LeakyReluSlope), name="e1")
#Network = tflearn.layers.normalization.batch_normalization(Network)
#current size: 256x256x3
#Network = tflearn.layers.merge_ops.merge((NetworkIN, Network), mode='concat', axis=3)
split0, _,_ = tf.split(NetworkIN, num_or_size_splits=3, axis=3)
Network = tflearn.layers.merge_ops.merge((split0, Network), mode='concat', axis=3)
#current size: 256x256x6
Network = tflearn.layers.conv.conv_2d(Network, 3, 4, strides=1, activation='tanh')
#Network = tflearn.layers.normalization.batch_normalization(Network)
Network = tf.nn.l2_normalize(Network, dim = 3)
#Network = regression(Network, optimizer='adam', learning_rate=LR, loss ='normal_loss_Masked', name = 'targets', batch_size=10, metric =None)
return tflearn.DNN(Network)