forked from bemc22/DeepLDSIR
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathAlpha8L.py
171 lines (141 loc) · 6.67 KB
/
Alpha8L.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
import tensorflow as tf
import keras
from tensorflow.keras.layers import Conv2D, Input, Lambda, Subtract
from tensorflow.keras.models import Model
from utils import spacial_tv2, dd_cassi, tf_dwt
class multiplyLayer(tf.keras.layers.Layer):
def __init__(self, name='Layer1'):
super(multiplyLayer, self).__init__(name=name)
def build(self, input_shape):
self.kernel = self.add_weight("kernel", shape=(
input_shape[1], input_shape[2], input_shape[3]),
initializer=tf.keras.initializers.glorot_normal,
trainable=True)
def call(self, input):
return tf.multiply(input, self.kernel)
class HSI_net:
def __init__(self, coded_aperture,
pretrained_weights=None, input_size=(512, 512, 31),
feature=64, denoiser="spacial_tv2"):
self.regul_term = {
"spacial_tv2": spacial_tv2
}
self.coded_aperture = coded_aperture
self.pretrained_weights = pretrained_weights
self.input_size = input_size
self.denoiser_dims = (1,) + input_size
self.encode_size = (input_size[0:2]) + (int(feature/8),)
self.krnl_regul = tf.keras.regularizers.L2(1e-8)
self.feature = feature
self.L = input_size[2]
self.denoiser_fun = self.regul_term[denoiser]
self.autoencoder = self.get_autoencoder()
self.encoder = self.get_encoder()
self.decoder = self.get_decoder()
self.modelR = self.get_modelR(self.denoiser_fun)
self.priorA = self.get_prior_autoencoder(self.denoiser_fun)
if(pretrained_weights):
for layer in self.autoencoder.layers[7:13]:
self.decoder.get_layer(layer.name).set_weights(
self.autoencoder.get_layer(layer.name).get_weights())
for layer in self.encoder.layers[1:7]:
self.encoder.get_layer(layer.name).set_weights(
self.autoencoder.get_layer(layer.name).get_weights())
for layer in self.modelR.layers:
if isinstance(layer, keras.layers.Conv2D):
self.modelR.get_layer(layer.name).set_weights(
self.autoencoder.get_layer(layer.name).get_weights())
layer.trainable = False
def E(self, inputs):
feature, krnl_regul = self.feature, self.krnl_regul
conv1 = Conv2D(feature, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv1')(inputs)
conv2 = Conv2D(feature, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv2')(conv1)
conv3 = Conv2D(feature/2, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv3')(conv2)
conv4 = Conv2D(feature/2, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv4')(conv3)
conv5 = Conv2D(feature/4, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv5')(conv4)
Eh = Conv2D(feature/8, 3, activation=None, padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='encode')(conv5)
return Eh
def D(self, inputs):
feature, krnl_regul = self.feature, self.krnl_regul
conv6 = Conv2D(feature/4, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv6')(inputs)
conv7 = Conv2D(feature/2, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv7')(conv6)
conv8 = Conv2D(feature/2, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv8')(conv7)
conv9 = Conv2D(feature, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv9')(conv8)
conv10 = Conv2D(feature, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='conv10')(conv9)
Dh = Conv2D(self.L, 3, activation='relu', padding='same',
kernel_initializer='glorot_normal',
kernel_regularizer=krnl_regul,
name='decode')(conv10)
return Dh
def get_autoencoder(self):
inputs = Input(self.input_size)
Eh = self.E(inputs)
DEh = self.D(Eh)
model = Model(inputs, DEh, name='Autoencoder')
if (self.pretrained_weights):
model.load_weights(self.pretrained_weights)
return model
def get_prior_autoencoder(self, denoiser):
inputs = Input(self.input_size)
Eh = self.E(inputs)
DEh = self.D(Eh)
TV = Lambda(lambda x: denoiser(x), name='TV')(DEh)
model = Model(inputs, [DEh, TV], name='PriorAutoencoder')
if (self.pretrained_weights):
model.load_weights(self.pretrained_weights)
return model
def get_encoder(self):
inputs = Input(self.input_size)
Eh = self.E(inputs)
model = Model(inputs, Eh, name='Encoder')
return model
def get_decoder(self):
inputs = Input(self.encode_size)
DEh = self.D(inputs)
model = Model(inputs, DEh, name='Decoder')
return model
def get_modelR(self, denoiser):
inputs = Input(self.encode_size, name="Input", batch_size=1)
layer = multiplyLayer(name='Layer1')(inputs)
decode = self.D(layer)
Iest = Lambda(lambda x: dd_cassi(
x, self.coded_aperture), name="I")(decode)
encode = self.E(decode)
P = Subtract(name='P')([layer, encode])
T = Lambda(lambda x: denoiser(x), name='TV')(decode)
T2 = Lambda(lambda x: tf_dwt(x), name='W')(layer)
model = Model(inputs, [Iest, P, T, T2], name='ReconsNet')
return model