Table 3.

Performance comparison on ETP dataset.

Methods	Backbone	Acc	F1
ACEDSNet (Li et al., 2021)	MobileNet	96.8	–
ResNet-50 + DA (Huang et al., 2022)	ResNet-50	97.1	96.1
Pre-trained DenseNet-169* (Huang et al., 2017)	DenseNet-169	99.2	99.2
Pre-trained ResNet-50* (He et al., 2016)	ResNet-50	99.7	99.7
Ours (general branch)	ResNet-50	99.7	99.7
Ours (improving branch)	ResNet-50	100	100

The bolded lines are the results obtained by our method, to emphasize.