Table 2.

Performance comparison on D0 dataset.

Methods	Backbone	Acc	F1
Pre-trained ResNet-50* (He et al., 2016)	ResNet-50	99.4	99.4
Pre-trained DenseNet-169* (Huang et al., 2017)	DenseNet-169	99.6	99.6
MMAL* (Zhang et al., 2021)	ResNet-50	99.8	99.8
Ours (general branch)	ResNet-50	99.8	99.8
CNNs Ensemble + Exp + ExpLR (Nanni et al., 2022)	EfficientNetB0 + ResNet-50 + GoogleNet + ShuffleNet + MobileNetV2 + DenseNet-201	99.8	99.7
MMALNet + DNVT + ResNet-50 + Ensemble (Xia et al., 2023)	ResNet-50 + DenseNet-201 + Transformer	99.9	99.9
Ours (improving branch)	ResNet-50	100	100

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