Self-Attention Multilayer Feature Fusion Based on Long Connection

Table 3

Comparative experiment.

Model	The calculation amount (B)	Acc. (%)
		Cifar10	Cifar100	Tiny-ImageNet
ResNet50 [7]	0.328	92.39	68.14	54.84
ResNet50-SCFF	0.362	93.20	74.27	60.78
ResNext50 [18]	0.340	92.78	70.84	54.80
ResNext50-SCFF	0.374	93.35	75.79	62.08
Res2Net50 [8]	0.342	92.58	70.85	57.06
Res2Net50-SCFF	0.375	93.37	74.31	61.29
PreActResNet50 [19]	0.328	93.35	73.08	54.74
PreActResNet50-SCFF	0.362	93.98	76.14	54.54
SKNet50 [9]	0.622	92.59	67.36	57.22
SKNet-SCFF	0.656	93.64	72.56	61.34
VGG16 [16]	0.333	92.66	—	—
VGG16-SCFF	0.339	93.30	—	—