qqpoll56188

It is well known that featuremap attention and multi-path representation are important for visual recognition. In this paper, we present a modularized