Hausdorff point convolution with geometric priors

作     者：Liqiang LIN Pengdi HUANG Fuyou XUE Kai XU Daniel COHEN-OR Hui HUANG Liqiang LIN;Pengdi HUANG;Fuyou XUE;Kai XU;Daniel COHEN-OR;Hui HUANG

作者机构：College of Computer Science and Software Engineering Shenzhen University School of Computer Science National University of Defense Technology 

出 版 物：《Science China(Information Sciences)》 (中国科学:信息科学(英文版))

年 卷 期：2021年第64卷第11期

页      面：67-79页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 080203[工学-机械设计及理论] 0802[工学-机械工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：supported by National Natural Science Foundation of China (Grant Nos. U2001206,61902254) Guangdong Talent Program (Grant No. 2019JC05X328) Guangdong Science and Technology Program (Grant Nos.2020A0505100064, 2015A030312015) DEGP Key Project (Grant Nos. 2018KZDXM058, 2020SFKC059) Shenzhen Science and Technology Program (Grant No. RCJC20200714114435012) National Engineering Laboratory for Big Data System Computing Technology Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ) 

主　　题：point convolution Hausdorff distance geometric prior deep neural network 

摘      要：Developing point convolution for irregular point clouds to extract deep features remains challenging. Current methods evaluate the response by computing point set distances which account only for the spatial alignment between two point sets, but not quite for their underlying shapes. Without a shapeaware response, it is hard to characterize the 3 D geometry of a point cloud efficiently with a compact set of kernels. In this paper, we advocate the use of modified Hausdorff distance as a shape-aware distance measure for calculating point convolutional responses. The technique we present, coined Hausdorff point convolution(HPC), is shape-aware. We show that HPC constitutes a powerful point feature learning with a rather compact set of only four types of geometric priors as kernels. We further develop an HPC-based deep neural network(HPC-DNN). Task-specific learning can be achieved by tuning the network weights for combining the shortest distances between the input and the kernel point sets. We also realize hierarchical feature learning by designing a multi-kernel HPC for multi-scale feature encoding. Extensive experiments demonstrate that HPC-DNN outperforms strong point convolution baselines(e.g., KPConv), achieving 2.8%m Io U performance boost on S3 DIS and 1.5% on Semantic KITTI for semantic segmentation task.