Zhang Wenxiao ; Xiao Chunxia

Abstract

Point cloud based retrieval for place recognition is an emerging problem invision field. The main challenge is how to find an efficient way to encode thelocal features into a discriminative global descriptor. In this paper, wepropose a Point Contextual Attention Network (PCAN), which can predict thesignificance of each local point feature based on point context. Our networkmakes it possible to pay more attention to the task-relevent features whenaggregating local features. Experiments on various benchmark datasets show thatthe proposed network can provide outperformance than current state-of-the-artapproaches.