Table 6.

Comparison of ResNet, DenseNet, MobileNetV2, and DeepLabv3+ as backbone.

Methods	Drishti-GS dataset			RIM-ONE-r3 dataset
	DSC cup	DSC disc	δ	DSC cup	DSC disc	δ
ResNet (He et al., 2016)	0.849	0.962	0.089	0.875	0.895	0.083
DenseNet (Huang et al., 2017)	0.831	0.954	0.122	0.782	0.892	0.126
MobileNetV2 (Sandler et al., 2018)	0.834	0.948	0.099	0.772	0.880	0.102
DeepLabv3+ (Chen et al., 2018)	0.857	0.962	0.083	0.791	0.898	0.082

The best result of the column is shown in bold.