Table 3.

Comparison of the baseline and proposed models on the test sets of the small- and large-scale image datasets in terms of loss and accuracy*.

Datasets	Models	Loss	AS	F1
CIFAR-10	Res2Net	1.155 ± 0.073	0.788 ± 0.009	0.760 ± 0.011
	LDR	1.694 ± 0.085	0.688 ± 0.013	0.659 ± 0.010
	ESM	1.040 ± 0.017	0.775 ± 0.010	0.756 ± 0.007
	VIN	2.070 ± 0.123	0.760 ± 0.012	0.715 ± 0.017
	ShuffleNetV2	1.220 ± 0.029	0.755 ± 0.015	0.706 ± 0.018
	MnasNet	2.588 ± 0.141	0.611 ± 0.034	0.597 ± 0.040
	MobileNetV3	1.479 ± 0.053	0.731 ± 0.012	0.692 ± 0.016
	Ours	1.146 ± 0.066	0.794 ± 0.023	0.769 ± 0.019
STL-10	Res2Net	1.948 ± 0.069	0.620 ± 0.011	0.600 ± 0.016
	LDR	2.001 ± 0.073	0.612 ± 0.019	0.587 ± 0.022
	ESM	1.678 ± 0.013	0.658 ± 0.014	0.644 ± 0.010
	VIN	1.530 ± 0.093	0.735 ± 0.010	0.716 ± 0.008
	ShuffleNetV2	1.895 ± 0.073	0.658 ± 0.007	0.643 ± 0.011
	MnasNet	2.163 ± 0.199	0.520 ± 0.018	0.504 ± 0.014
	MobileNetV3	1.723 ± 0.036	0.669 ± 0.010	0.661 ± 0.08
	Ours	1.137 ± 0.024	0.735 ± 0.007	0.723 ± 0.010
ImageNet-100	Res2Net	0.863 ± 0.083	0.835 ± 0.009	0.757 ± 0.006
	LDR	0.896 ± 0.090	0.852 ± 0.011	0.775 ± 0.013
	ESM	0.837 ± 0.102	0.874 ± 0.016	0.809 ± 0.018
	VIN	0.894 ± 0.067	0.841 ± 0.007	0.781 ± 0.006
	ShuffleNetV2	0.902 ± 0.049	0.847 ± 0.010	0.777 ± 0.008
	MnasNet	1.431 ± 0.093	0.744 ± 0.014	0.652 ± 0.017
	MobileNetV3	1.312 ± 0.048	0.762 ± 0.005	0.654 ± 0.007
	Ours	0.852 ± 0.013	0.863 ± 0.012	0.801 ± 0.013

Lower loss and higher AS and F1 scores correspond to better performance of a model. *This information is based on experiments using 32 GB NVIDIA Tesla V100-SXM2 GPU

Significant values are in [bold].