-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
170 lines (142 loc) · 7.8 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
'''
Copyright (C) 2020 Vignesh Kotteeswaran
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
'''
from __future__ import division
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()
import numpy as np
from tensorflow.keras.models import Model
from tensorflow.keras.layers import Input, Lambda, Conv2D, SeparableConv2D, MaxPooling2D, BatchNormalization, ELU, Reshape, Concatenate, Activation,Add,GlobalAveragePooling2D,Multiply
from tensorflow.keras.regularizers import l2
import tensorflow.keras.backend as K
import utils
def get_shape(z):
if hasattr(z,'_keras_shape'):
#print(z._keras_shape[-1])
return z._keras_shape[-1]
elif hasattr(z,'shape'):
#if hasattr(x.shape,'value'):
#print(z.shape[-1].value)
return z.shape[-1].value
def se(x_):
y=GlobalAveragePooling2D()(x_)
y=Reshape(target_shape=(1,1,int(get_shape(y))))(y)
y=Conv2D(int(get_shape(y)*0.25),(1,1),padding='same')(y)
y=swish(y)
y=Conv2D(int(get_shape(x_)),(1,1),padding='same')(y)
y=Activation('sigmoid')(y)
y=Multiply()([x_,y])
y=Conv2D(int(get_shape(y)),(1,1),padding='same')(y)
y=swish(y)
return y
def swish(x_):
#return Lambda(swish_fn, output_shape=x_._keras_shape[1:])(x_)
#return Multiply()([x_,Activation('tanh')(Activation('softplus')(x_))])
#return Multiply()([x_,Activation('sigmoid')(x_)])
return Activation('relu')(x_)
def build_model(image_size,
redux=1.0,
l2_regularization=0.0,
show_flops=False,
first_stride=2):
l2_reg = l2_regularization
img_height, img_width, img_channels = image_size[0], image_size[1], image_size[2]
############################################################################
# Build the network.
############################################################################
x = Input(shape=(img_height, img_width, img_channels))
conv1 = Conv2D(int(16*redux), (3,3), strides=(first_stride, first_stride), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv1')(x)
bn1 = BatchNormalization(axis=3, momentum=0.99, name='bn1')(conv1) # Tensorflow uses filter format [filter_height, filter_width, in_channels, out_channels], hence axis = 3
#conv1 = ELU(name='elu1')(conv1)
act1=se(swish(bn1))
#add1=Add()([conv1,act1])
#bn1_a = BatchNormalization(axis=3, momentum=0.99, name='bn1_a')(add1)
#add_act1=swish(bn1_a)
pool1 = MaxPooling2D(pool_size=(2, 2), name='pool1')(act1)
'''
conv2 = Conv2D(48, (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv2')(add_act1)
bn2 = BatchNormalization(axis=3, momentum=0.99, name='bn2')(conv2)
#conv2 = ELU(name='elu2')(conv2)
act2=Activation('relu')(bn2)
add2=Add()([conv2,act2])
bn2_a = BatchNormalization(axis=3, momentum=0.99, name='bn2_a')(add2)
add_act2=Activation('relu')(bn2_a)
#pool2 = MaxPooling2D(pool_size=(2, 2), name='pool2')(add_act2)
'''
conv3 = Conv2D(int(16*redux), (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv3')(pool1)
#conv3=SeparableConv2D(filters=int(64*redux),kernel_size=(3,3),padding='same',depthwise_initializer='he_normal', pointwise_initializer='he_normal')(pool1)
bn3 = BatchNormalization(axis=3, momentum=0.99, name='bn3')(conv3)
#conv3 = ELU(name='elu3')(conv3)
act3=swish(bn3)
add3=Add()([pool1,act3])
bn3_a = BatchNormalization(axis=3, momentum=0.99, name='bn3_a')(add3)
add_act3=se(swish(bn3_a))
pool3 = MaxPooling2D(pool_size=(2, 2), name='pool3')(add_act3)
conv4 = Conv2D(int(32*redux), (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv4')(pool3)
#conv4=SeparableConv2D(filters=int(64*redux),kernel_size=(3,3),padding='same',depthwise_initializer='he_normal', pointwise_initializer='he_normal')(pool3)
bn4 = BatchNormalization(axis=3, momentum=0.99, name='bn4')(conv4)
#conv4 = ELU(name='elu4')(conv4)
act4=swish(bn4)
#add4=Add()([pool3,act4])
#bn4_a = BatchNormalization(axis=3, momentum=0.99, name='bn4_a')(add4)
#add_act4=swish(bn4_a)
act4=se(act4)
pool4 = MaxPooling2D(pool_size=(2, 2), name='pool4')(act4)
conv5 = Conv2D(int(32*redux), (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv5')(pool4)
#conv5=SeparableConv2D(filters=int(48*redux),kernel_size=(3,3),padding='same',depthwise_initializer='he_normal', pointwise_initializer='he_normal')(pool4)
bn5 = BatchNormalization(axis=3, momentum=0.99, name='bn5')(conv5)
#conv5 = ELU(name='elu5')(conv5)
act5=swish(bn5)
add5=Add()([pool4,act5])
bn5_a = BatchNormalization(axis=3, momentum=0.99, name='bn5_a')(add5)
add_act5=swish(bn5_a)
add_act5=se(add_act5)
pool5 = MaxPooling2D(pool_size=(2, 2), name='pool5')(add_act5)
conv6 = Conv2D(int(64*redux), (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv6')(pool5)
#conv6=SeparableConv2D(filters=int(48*redux),kernel_size=(3,3),padding='same',depthwise_initializer='he_normal', pointwise_initializer='he_normal')(pool5)
bn6 = BatchNormalization(axis=3, momentum=0.99, name='bn6')(conv6)
#conv6 = ELU(name='elu6')(conv6)
act6=swish(bn6)
#add6=Add()([pool5,act6])
#bn6_a = BatchNormalization(axis=3, momentum=0.99, name='bn6_a')(add6)
#add_act6=swish(bn6_a)
act6=se(act6)
pool6 = MaxPooling2D(pool_size=(2, 2), name='pool6')(act6)
conv7 = Conv2D(int(64*redux), (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv7')(pool6)
#conv7=SeparableConv2D(filters=int(32*redux),kernel_size=(3,3),padding='same',depthwise_initializer='he_normal', pointwise_initializer='he_normal')(pool6)
bn7 = BatchNormalization(axis=3, momentum=0.99, name='bn7')(conv7)
#conv7 = ELU(name='elu7')(conv7)
act7=swish(bn7)
add7=Add()([pool6,act7])
bn7_a = BatchNormalization(axis=3, momentum=0.99, name='bn7_a')(add7)
add_act7=swish(bn7_a)
add_act7=se(add_act7)
'''
conv8 = Conv2D(int(64*redux), (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv8')(add_act7)
#conv7=SeparableConv2D(filters=int(32*redux),kernel_size=(3,3),padding='same',depthwise_initializer='he_normal', pointwise_initializer='he_normal')(pool6)
bn8 = BatchNormalization(axis=3, momentum=0.99, name='bn8')(conv8)
#conv7 = ELU(name='elu7')(conv7)
act8=swish(bn8)
add8=Add()([conv7,act8])
bn8_a = BatchNormalization(axis=3, momentum=0.99, name='bn8_a')(add8)
add_act8=swish(bn8_a)
add_act8=se(add_act8)
conv9 = Conv2D(int(128*redux), (3, 3), strides=(1, 1), padding="same", kernel_initializer='he_normal', kernel_regularizer=l2(l2_reg), name='conv9')(add_act8)
#conv7=SeparableConv2D(filters=int(32*redux),kernel_size=(3,3),padding='same',depthwise_initializer='he_normal', pointwise_initializer='he_normal')(pool6)
bn9 = BatchNormalization(axis=3, momentum=0.99, name='bn9')(conv9)
act9=swish(bn9)
add_act9=se(act9)
'''
model = Model(inputs=x, outputs=add_act7)
if show_flops:
utils.show_flops(model,image_size)
return model