Table 1 Backbone realization details.

From: Object detection model design for tiny road surface damage

Stage

Operator module

Input size

Output size

s

k

d

n

0

Input image

3 × 640 × 640

1

DSCD

3 × 640 ×  640

32 × 320 × 320

2

7

1

DSCD

32 × 320 ×  320

64 × 160 × 160

2

7

1

MFE_SCR

64 × 160 ×  160

64 × 160 × 160

1

1

2

DSCD

64 × 160 ×  160

128 × 80 × 80

2

5

1

MFE_PCR

128 × 80 ×  80

128 × 80 × 80

1

1,2,3

1

3

DSCD

128 × 80 ×  80

256 × 40 × 40

2

5

1

MFE_PCR

256 × 40 ×  40

256 × 40 × 40

1

1,3,5

1

4

DSCD

256 × 40 ×  40

512 × 20 × 20

2

5

1

MFE_PCR

512 × 20 ×  20

512 × 20 × 20

1

1,3,5

1

SPPF

512 × 20 ×  20

512 × 20 × 20

1

  1. In the table, “s” represents the stride of the first convolution in each module; “k” represents the kernel size of the second convolution in DSCD; “d” represents the dilation rate of parallel convolution in MFE_PCR; and “n” represents the number of modules.