Table 2 Architecture details of the DilatedSkinNet

Input layer

\(192\times 256\times 3\)

–

–

\(192\times 256\times 3\)

Block 1

Conv1

\(3\times 3\), 8

1

\(192\times 256\times 8\)

Block 2

Conv2_1

\(3\times 3\), 16

2

\(192\times 256\times 16\)

Conv2_2

\(3\times 3\), 16

4

\(192\times 256\times 16\)

Block 3

Conv3_1

\(3\times 3\), 32

4

\(192\times 256\times 32\)

Conv3_2

\(1\times 1\), 16

6

\(192\times 256\times 16\)

Conv3_3

\(3\times 3\), 32

8

\(192\times 256\times 32\)

Block 4

Conv4_1

\(3\times 3\), 64

8

\(192\times 256\times 64\)

Conv4_2

\(1\times 1\), 32

10

\(192\times 256\times 32\)

Conv4_3

\(3\times 3\), 64

10

\(192\times 256\times 64\)

Conv4_4

\(3\times 3\), 64

12

\(192\times 256\times 64\)

Block 5

Conv5_1

\(3\times 3\), 128

12

\(192\times 256\times 128\)

Conv5_2

\(1\times 1\), 64

12

\(192\times 256\times 64\)

Conv5_3

\(3\times 3\), 128

14

\(192\times 256\times 128\)

Conv5_4

\(1\times 1\), 64

14

\(192\times 256\times 64\)

Conv5_5

\(3\times 3\), 128

14

\(192\times 256\times 128\)

final_Conv

\(1\times 1\), 2

–

\(192\times 256\times 2\)

Softmax

–

–

\(192\times 256\times 2\)

OutPutMap

Pixel classification

Cross entropy

Loss function

\(192\times 256\times 2\)