Multiactivation Pooling Method in Convolutional Neural Networks for Image Recognition

Table 3

Error rates in the test dataset of CIFAR-10. Top: error rates based on VGG-11 model and its extension models. Bottom: error rates based on VGG-13 model and its extension models.

	VGG-11 error rate (%)
Baseline model(model A)	8.72
With large-scale pooling method	44 maxpool; (model B)	88 maxpool; (model C)	1616 maxpool; (model D)
	8.24	7.84	8.19
With MAP method	44; MAP; (model B)	88; MAP; (model C)	1616 MAP; (model D)
	8.07	6.94	7.76
	VGG-13 error rate (%)
Baseline model(model E)	7.50
With large-scale pooling method	44 maxpool; (model F)	88 maxpool; (model G)	1616 maxpool; (model H)
	7.08	7.22	7.35
With MAP method	44; MAP; (model F)	88; MAP; (model G)	1616 MAP; (model H)
	6.96	6.85	7.07