Automated Pavement Crack Damage Detection Using Deep Multiscale Convolutional Features

Table 1

Details of upsampling module units.

Items	Kernel size	Numbers of feature maps	Feature size
Input	—	—	H, W
Low-level features	1 × 1	256	H/4, W/4
High-level features	1 × 1	512	H/16, W/16
Conv-1	3 × 3	256	H/4, W/4
Conv-2	3 × 3	256	H/4, W/4
Conv-3	3 × 3	256	H/4, W/4
Output	1 × 1	2	H, W