Age Estimation Using Channel Aggregation Transform Based On Deep Neural Network

Abstract

With the rapid development of deep learning, the accuracy of models is getting higher and higher, but it is difficult to balance the interpretability and accuracy of deep network. This paper proposes a modular aggregation-attention module, which has the same topological structure. After channel grouping, channel level information is exchanged through channel level attention, and finally, a new NDF variant CA-NEXT is obtained by combining with NDF. We provide detailed empirical data and the resulting model accuracy can improve the accuracy.