-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodel.py
175 lines (144 loc) · 8.39 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
# ----------------------------------------------Import required Modules----------------------------------------------- #
import tensorflow as tf
from tensorflow.keras.applications import VGG16, ResNet50, DenseNet121
# ----------------------------------------------Define Model---------------------------------------------------------- #
# Build complete autoencoder model
def build_autoencoder(input_shape = (224, 224, 3), enc_net = "vgg", describe = False):
'''
Build Autoencoder Model.\n
:param input_shape: Input Shape passed to Autoencoder Model (224,224,3) (default)\n
:return: Autoencoder Model
'''
def encoder(inp, enc_net = "vgg", input_shape=(224,224,3)):
'''
Build Encoder Model.\n
:param inp: Input to Autoencoder Model\n
:param input_shape: Input Shape passed to Autoencoder Model (224,224,3) (default)\n
:return: Encoder Model
'''
if enc_net == "vgg":
cnn_model = VGG16(include_top = False,
weights = "imagenet",
input_shape = input_shape,
pooling = "none")
cnn_model.trainable = False
pre_trained = tf.keras.models.Model(inputs = cnn_model.input,
outputs = cnn_model.get_layer(name="block4_conv2").output,
name = "vgg")
elif enc_net == "resnet":
cnn_model = ResNet50(include_top = False,
weights = "imagenet",
input_shape = input_shape,
pooling = "none")
cnn_model.trainable = False
pre_trained = tf.keras.models.Model(inputs = cnn_model.input,
outputs = cnn_model.get_layer(name="conv3_block1_out").output,
name = "resnet")
elif enc_net == "densenet":
cnn_model = DenseNet121(include_top = False,
weights = "imagenet",
input_shape = input_shape,
pooling = "none")
cnn_model.trainable = False
pre_trained = tf.keras.models.Model(inputs = cnn_model.input,
outputs = cnn_model.get_layer(name="pool3_relu").output,
name = "densenet")
# https://keras.io/guides/transfer_learning/
x = pre_trained(inputs = inp, training=False)
# print(pre_trained.summary())
layer10 = tf.keras.layers.Conv2D(filters = 512,
kernel_size = 1,
name = "conv10")(x) # for pix2vox-A(large), kernel_size is 3
layer10_norm = tf.keras.layers.BatchNormalization(name="layer10_norm")(layer10)
layer10_elu = tf.keras.activations.elu(layer10_norm,
name="layer10_elu")
layer11 = tf.keras.layers.Conv2D(filters = 256,
kernel_size = 3,
name = "conv11")(layer10_elu) # for pix2vox-A(large), filters is 512
layer11_norm = tf.keras.layers.BatchNormalization(name="layer11_norm")(layer11)
layer11_elu = tf.keras.activations.elu(layer11_norm,
name="layer11_elu")
layer11_pool = tf.keras.layers.MaxPooling2D(pool_size = (4,4),
name="layer11_pool")(layer11_elu) # for pix2vox-A(large), kernel size is 3
layer12 = tf.keras.layers.Conv2D(filters = 128,
kernel_size = 3,
name = "conv12")(layer11_pool) # for pix2vox-A(large), filters is 256, kernel_size is 1
layer12_norm = tf.keras.layers.BatchNormalization(name="layer12_norm")(layer12)
layer12_elu = tf.keras.activations.elu(layer12_norm,
name="layer12_elu")
return layer12_elu
def decoder(inp):
'''
Build Decoder Model.\n
:param inp: Reshaped Output of Encoder Model\n
:return: Decoder Model
'''
layer1 = tf.keras.layers.Convolution3DTranspose(filters=128,
kernel_size=4,
strides=(2,2,2),
padding="same",
use_bias=False,
name="Conv3D_1")(inp)
layer1_norm = tf.keras.layers.BatchNormalization(name="layer1_norm")(layer1)
layer1_relu = tf.keras.activations.relu(layer1_norm,
name="layer1_relu")
layer2 = tf.keras.layers.Convolution3DTranspose(filters=64,
kernel_size=4,
strides=(2,2,2),
padding="same",
use_bias=False,
name="Conv3D_2")(layer1_relu)
layer2_norm = tf.keras.layers.BatchNormalization(name="layer2_norm")(layer2)
layer2_relu = tf.keras.activations.relu(layer2_norm,
name="layer2_relu")
layer3 = tf.keras.layers.Convolution3DTranspose(filters=32,
kernel_size=4,
strides=(2,2,2),
padding="same",
use_bias=False,
name="Conv3D_3")(layer2_relu)
layer3_norm = tf.keras.layers.BatchNormalization(name="layer3_norm")(layer3)
layer3_relu = tf.keras.activations.relu(layer3_norm,
name="layer3_relu")
layer4 = tf.keras.layers.Convolution3DTranspose(filters=8,
kernel_size=4,
strides=(2,2,2),
padding="same",
use_bias=False,
name="Conv3D_4")(layer3_relu)
layer4_norm = tf.keras.layers.BatchNormalization(name="layer4_norm")(layer4)
layer4_relu = tf.keras.activations.relu(layer4_norm,
name="layer4_relu")
layer5 = tf.keras.layers.Convolution3DTranspose(filters=1,
kernel_size=1,
use_bias=False,
name="Conv3D_5")(layer4_relu)
layer5_sigmoid = tf.keras.activations.sigmoid(layer5,
name="layer5_sigmoid")
# TODO: check this statement
layer5_sigmoid = tf.keras.layers.Reshape((32,32,32))(layer5_sigmoid)
return layer5_sigmoid
# Input
input = tf.keras.Input(shape = input_shape, name = "input_layer")
# Encoder Model
encoder_model = tf.keras.Model(input, encoder(input, enc_net), name = "encoder")
if describe:
print("\nEncoder Model Summary:\n")
encoder_model.summary()
# Decoder Input Reshaped from Encoder Output
decoder_input = tf.keras.Input(shape=(2, 2, 2, 256),
name = "decoder_input")
# Decoder Model
decoder_model = tf.keras.Model(decoder_input, decoder(decoder_input), name = "decoder")
if describe:
print("\nDecoder Model Summary:\n")
decoder_model.summary()
# Autoencoder Model
encoder_output = encoder_model(input)
# the encoder output should be reshaped to (-1,2,2,2,256) to be fed into decoder
decoder_input = tf.keras.layers.Reshape((2,2,2,256))(encoder_output)
autoencoder_model = tf.keras.Model(input, decoder_model(decoder_input), name ='autoencoder')
if describe:
print("\nAutoencoder Model Summary:\n")
autoencoder_model.summary()
return autoencoder_model