Multiactivation Pooling Method in Convolutional Neural Networks for Image Recognition

Table 4

Error rates in the test dataset of CIFAR-100, SVHN, and ImageNet on VGG models.

Models	CIFAR-100 (%)	SVHN (%)	ImageNet top-5 (%)
Baseline model(A)	30.96	3.04	10.86
With 44 large-scale pooling method(B)	30.37	2.87	-
With 44 MAP method(B)	30.04	2.44	-
With 88 large-scale pooling method(C)	29.45	2.36	10.54
With 88 MAP method(C)	28.37	2.03	10.19
Baseline model(E)	29.05	2.27	-
With 44 large-scale pooling method(F)	28.62	2.14	-
With 44 MAP method(F)	27.92	1.98	-
With 88 large-scale pooling method(G)	28.03	2.05	-
With 88 MAP method(G)	27.4	1.89	-