[Retracted] Channel-Wise Correlation Calibrates Attention Module for Convolutional Neural Networks

Table 5

Test accuracy (%) on different models on mini-ImageNet.

Model	Para (M)	GFloat	Top1	Top5
DenseNet-121	8.081	2.898	82.45	95.17
DenseNet-121+SE	8.113	2.898	82.61	95.17
DenseNet-121+nonlocal	8.084	2.899	82.19	95.09
DenseNet-121+CBAM	8.119	2.902	80.60	94.56
DenseNet-121+GC	8.113	2.899	79.47	93.94
DenseNet-121+LCM	8.112	2.898	83.05	95.23
ResNet-50	23.71	4.132	80.54	94.60
ResNet-50+SE	26.24	4.137	81.46	94.94
ResNet-50+CBAM	26.25	4.143	82.33	95.15
ResNet-50+nonlocal	25.82	4.544	80.23	94.23
ResNet-50+GC	26.24	4.138	77.06	92.89
ResNet-50+LCM	24.97	4.133	81.94	95.09
ResNeXt-50 32 × 4d	23.19	4.287	81.54	94.76
ResNeXt-50 32 × 4d + SE	25.72	4.292	82.40	95.06
ResNeXt-50 32 × 4d + CBAM	25.72	4.298	82.56	95.31
ResNeXt-50 32 × 4d + nonlocal	25.29	4.699	81.13	94.53
ResNeXt-50 32 × 4d + GC	33.19	4.300	76.18	92.26
ResNeXt-50 32 × 4d + LCM	23.81	4.287	82.72	95.25