A Channel-aware Attention Network for Crowd Counting

Abstract

With the rapid increase of urban population, crowd counting is a popular yet difficult topic. However, the problem of scale variation in high-density scenario remains under-explored. To address this problem, we propose a channel-aware attention network in this paper. The channel attention module attempts to handle the relations between channel maps and highlight the discriminative information in specific channels. Thus, it alleviates the misestimation for background regions. Experimental results on ShanghaiTech and UCF-QNRF benchmark datasets prove that our approach achieves compelling performance compared to the state-of-the-art methods.