Semantic_Segmentation_Models/DeepLabV3plus_SqueezeNet.py at main · tshr-d-dragon/Semantic_Segmentation_Models · GitHub

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
import os
os.environ["TF_CPP_MIN_LOG_LEVEL"] = "3"

from keras.layers import Conv2D, BatchNormalization, Activation, UpSampling2D, MaxPooling2D
from keras.layers import AveragePooling2D, Concatenate, Input, GlobalAveragePooling2D
from keras.models import Model


def SqueezeNet(input_shape):

    def Fire(inputs, fs, fe):

      s1 = Conv2D(filters=fs, kernel_size=1, padding='same', use_bias=False, activation='relu')(inputs)
      s1 = BatchNormalization()(s1)
      e1 = Conv2D(filters=fe, kernel_size=1, padding='same', use_bias=False, activation='relu')(s1)
      e1 = BatchNormalization()(e1)
      e3 = Conv2D(filters=fe, kernel_size=3, padding='same', use_bias=False, activation='relu')(s1)
      e3 = BatchNormalization()(e3)

      output = Concatenate()([e1, e3])

      return output


    x1 = Conv2D(filters=96, kernel_size=7, strides=2, padding='same', use_bias=False, activation='relu')(input_shape)
    x1= s1 = BatchNormalization()(x1)
    x1 = MaxPooling2D(pool_size=3, strides=2, padding='same')(x1)

    f2 = Fire(x1, 16, 64)
    f3 = Fire(f2, 16, 64)
    f4 = Fire(f3, 32, 128)
    x1 = MaxPooling2D(pool_size=3, strides=2, padding='same')(f4)

    f5 = Fire(x1, 32, 128)
    f6 = Fire(f5, 48, 192)
    f7 = Fire(f6, 48, 192)
    f8 = Fire(f7, 64, 256)
    x1 = MaxPooling2D(pool_size=3, strides=2, padding='same')(f8)

    f8 = Fire(x1, 64, 256)

    return f4, f8


def ASPP(image_features):

    shape = image_features.shape

    y_pool = AveragePooling2D(pool_size=(shape[1], shape[2]))(image_features)
    y_pool = Conv2D(filters=128, kernel_size=1, padding='same', use_bias=False)(y_pool)
    y_pool = BatchNormalization(name=f'bn_1')(y_pool)
    y_pool = Activation('relu', name=f'relu_1')(y_pool)
    y_pool = UpSampling2D((shape[1], shape[2]), interpolation="bilinear")(y_pool)

    y_1 = Conv2D(filters=128, kernel_size=1, padding='same', use_bias=False)(image_features)
    y_1 = BatchNormalization(name=f'bn_2')(y_1)
    y_1 = Activation('relu', name=f'relu_2')(y_1)

    y_6 = Conv2D(filters=128, kernel_size=3, padding='same', dilation_rate = 6,use_bias=False)(image_features)
    y_6 = BatchNormalization(name=f'bn_3')(y_6)
    y_6 = Activation('relu', name=f'relu_3')(y_6)

    y_12 = Conv2D(filters=128, kernel_size=3, padding='same', dilation_rate = 12,use_bias=False)(image_features)
    y_12 = BatchNormalization(name=f'bn_4')(y_12)
    y_12 = Activation('relu', name=f'relu_4')(y_12)

    y_18 = Conv2D(filters=128, kernel_size=3, padding='same', dilation_rate = 6,use_bias=False)(image_features)
    y_18 = BatchNormalization(name=f'bn_5')(y_18)
    y_18 = Activation('relu', name=f'relu_5')(y_18)

    y_c = Concatenate()([y_pool, y_1, y_6, y_12, y_18])

    y = Conv2D(filters=128, kernel_size=1, padding='same', use_bias=False)(y_c)
    y = BatchNormalization(name=f'bn_6')(y)
    y = Activation('relu', name=f'relu_6')(y)

    return y

def DeepLabV3Plus(inputs, classes=1):

    inputs = Input(inputs)

    low_level_image_features, high_level_image_features = SqueezeNet(input_shape = inputs)

    x_a = ASPP(high_level_image_features)
    x_a = UpSampling2D(size=4, interpolation='bilinear')(x_a)

    x_b = Conv2D(filters=128, kernel_size=1, padding='same', use_bias=False)(low_level_image_features)
    x_b = BatchNormalization(name=f'bn_7')(x_b)
    x_b = Activation('relu', name=f'relu_7')(x_b)

    x = Concatenate()([x_a, x_b])

    x = Conv2D(filters=128, kernel_size=3, padding='same', use_bias=False)(x)
    x = BatchNormalization(name=f'bn_8')(x)
    x = Activation('relu', name=f'relu_8')(x)

    x = Conv2D(filters=128, kernel_size=3, padding='same', use_bias=False)(x)
    x = BatchNormalization(name=f'bn_9')(x)
    x = Activation('relu', name=f'relu_9')(x)

    x = UpSampling2D(size=4, interpolation='bilinear')(x)

    x = Conv2D(classes, (1, 1), name='output_layer')(x)

    if classes == 1:
      x = Activation('sigmoid')(x)
    else:
      x = Activation('softmax')(x)

    model = Model(inputs=inputs, outputs=x)

    return model


def main():

    model = DeepLabV3Plus(inputs=(512,512,3), classes=1)
    model.summary()

if __name__== '__main__':

    main()