Improvement of Residual Attention Network for Image Classification

摘要

The existing residual attention network (RAN) method mainly utilizes the deeper network layer for the image objects which are to be classified. However, when the network depth is simply increased, it will lead to gradient dispersion (or explosion) effect. To address the problem, we propose a new improvement method of residual attention network for image classification, which applies several upsampling schemes to the RAN process, i.e., the stacked network structure extraction, and the bottom-up and top-down feedforward attention for residual learning. In the experiments, we have given comparisons of different network structures and different upsampling methods that are applied to the RAN. The proposed improvement method achieves state-of-the-art image classification performance on two benchmark datasets including CIFAR-10 (4.23% error) and CIFAR-100 (21.15% error). Compared with the traditional RAN method, the proposed improvement method can improve the accuracy of image classification to some extent.