Multiscale Road Extraction in Remote Sensing Images

Table 2

Comparison with U-Net, LinkNet, DeepLabv3+, and D-LinkNet on the Massachusetts Roads test datasets.

Model	Pixel accuracy	mIoU
U-Net	0.9566	0.5985
LinkNet	0.9678	0.6792
DeepLabv3+	0.9667	0.6672
D-LinkNet	0.9675	0.6703
Ours	0.9672	0.6878