TABLE III:

Performance comparison between the proposed method with state-of-the-art algorithms. Bolded numbers indicate the top performance across all models being compared.

Method	Test set 1	Test set 2	Test set 3
CE	0.924 ± 0.02	0.836 ± 0.01	0.585 ± 0.01
SCE	0.929 ± 0.02	0.843 ± 0.03	0.558 ± 0.01
Co-teaching	0.905 ± 0.01	0.824 ± 0.02	0.539 ± 0.01
INCV	0.932 ± 0.01	0.861 ± 0.01	0.605 ± 0.01
DivideMix	0.931 ± 0.01	0.891 ± 0.01	0.737 ± 0.01
ELR	0.860 ± 0.02	0.811 ± 0.01	0.566 ± 0.01
SOP	0.930 ± 0.02	0.887 ± 0.02	0.661 ± 0.01
CMC (Ours)	0.932 ± 0.02	0.910 ± 0.02	0.735 ± 0.01

Method (AUROC)	Test set 1		Test set 2		Test set 3
	Good quality	Bad quality	Good quality	Bad quality	Good quality	Bad quality
CE	0.943 ± 0.02	0.912 ± 0.02	0.934 ± 0.02	0.768 ± 0.02	0.896 ± 0.02	0.542 ± 0.02
SCE	0.935 ± 0.02	0.905 ± 0.02	0.929 ± 0.02	0.755 ± 0.02	0.710 ± 0.02	0.531 ± 0.02
Co-teaching	0.917 ± 0.01	0.875 ± 0.01	0.905 ± 0.01	0.731 ± 0.01	0.777 ± 0.01	0.500 ± 0.01
INCV	0.942 ± 0.01	0.927 ± 0.01	0.942 ± 0.01	0.778 ± 0.01	0.915 ± 0.01	0.547 ± 0.01
DivideMix	0.943 ± 0.01	0.926 ± 0.01	0.938 ± 0.01	0.819 ± 0.02	0.964 ± 0.01	0.695 ± 0.01
ELR	0.869 ± 0.02	0.861 ± 0.02	0.919 ± 0.02	0.684 ± 0.02	0.743 ± 0.02	0.524 ± 0.02
SOP	0.937 ± 0.02	0.914 ± 0.02	0.987 ± 0.02	0.797 ± 0.02	0.958 ± 0.02	0.571 ± 0.02
CMC (Ours)	0.937 ± 0.02	0.927 ± 0.02	0.991 ± 0.02	0.843 ± 0.01	0.974 ± 0.02	0.670 ± 0.02

^*Tested on the whole dataset

^*Tested on good and bad quality subgroups