Table IV.

The quantitative results of infection areas on our uAI30 dataset compared to the corresponding ground-truth using U-Net architecture [49] and VGG16 FCN8 network [57]. The best results are highlight in bold.

Method	backbone	DI	JA	SE	SP
WSCL [18]	U-Net	0.702	0.550	0.819	0.803
S2L [44]		0.732	0.589	0.856	0.927
USTM -Net (proposed)		0.751	0.616	0.867	0.971
WSCL [18]	FCN8	0.685	0.513	0.806	0.785
S2L [44]		0.720	0.575	0.820	0.909
USTM -Net (proposed)		0.734	0.592	0.854	0.949