Multiactivation Pooling Method in Convolutional Neural Networks for Image Recognition

Table 8

Error rates in the test dataset of CIFAR and SVHN. The first row shows the error rate of the baseline model DenseNet-40 from paper ‎[15]. The second row shows the error rate of our implement on the baseline model DenseNet-40. The third row shows the error rate of our DenseNet-40_MAP model in Table 7.

Method	CIFAR-10 (%)	CIFAR-100 (%)	SVHN (%)	Params
Densenet-40 ‎[15]	5.24	24.42	1.79	1.05M
Densenet-40_ours	5.43	24.91	1.92	1.05M
Densenet-40_MAP	4.99	24.03	1.68	4.32M