Table 1 The structure of the improved MobileNetV3 model.

From: A lightweight semantic segmentation method for concrete bridge surface diseases based on improved DeeplabV3+

Input

Operator

Expansion rate

#out

ECA-Net

AF

Stride

5122 × 3

conv,3 × 3

–

16

 

HS

2

2562 × 16

bneck,3 × 3

16

16

 

RE

1

2562 × 16

bneck,3 × 3

64

24

 

RE

2

1282 × 24

bneck,3 × 3

72

24

 

RE

1

1282 × 24

bneck,5 × 5

72

40

√

RE

2

642 × 40

bneck,5 × 5

120

40

√

RE

1

642 × 40

bneck,5 × 5

120

40

√

RE

1

642 × 40

bneck,3 × 3

240

80

 

HS

1

642 × 80

bneck,3 × 3

200

80

 

HS

1

642 × 80

bneck,3 × 3

184

80

 

HS

2

322 × 80

bneck,3 × 3

184

80

 

HS

1

322 × 80

bneck,3 × 3

480

112

√

HS

1

322 × 112

bneck,3 × 3

672

112

√

HS

1

322 × 112

bneck,5 × 5

672

160

√

HS

2

162 × 160

bneck,5 × 5

960

160

√

HS

1

162 × 160

bneck,5 × 5

960

160

√

HS

1

162 × 160

conv,1 × 1

–

1280

 

HS

1

  1. Note: bneck denotes the depthwise separable convolution module, #out denotes the number of output channels, AF denotes the type of activation function, HS denotes the h-swish activation function and RE denotes the ReLU activation function.
Back to article page