Table V.

Comparison of the State-of-the-art Methods and Our Proposed CA-NET For Skin Lesion Segmentation. INF-T Means the Inference Time for a Single Image. E-Able Means the Method Is Explainable

Network	Para/Inf-T	E-able	Dice(%)	ASSD(pix)
Baseline(U-Net [6])	1.9M/1.7ms	×	87.77±3.51	1.23±1.07
Attention U-Net [9]	2.1M/1.8ms	√	88.46±3.37	1.18±1.24
DenseASPP [35]	8.3M/4.2ms	×	90.80±3.81	0.59±0.70
DeepLabv3+(DRN)	40.7M/2.2ms	×	91.79±3.39	0.54±0.64
RefineNet [37]	46.3M/3.4ms	×	91.55±2.11	0.64±0.77
DeepLabv3÷ [20]	54.7M/4.0ms	×	92.21 ±3.38	0.48 ±0.58
CA-Net(Ours)	2.8M/2.1ms	√	92.08±2.67	0.58±0.56