Weakly supervised point cloud segmentation via deep morphological semantic information embedding

作     者：Wenhao Xue Yang Yang Lei Li Zhongling Huang Xinggang Wang Junwei Han Dingwen Zhang 

作者机构：School of AutomationNorthwestern Polytechnical UniversityXi'anChina Institute of Mechanical and Electrical EngineeringBeijingChina School of Electronic Information and CommunicationsHuazhong University of Science and TechnologyWuhanChina 

出 版 物：《CAAI Transactions on Intelligence Technology》 (智能技术学报（英文）)

年 卷 期：2024年第9卷第3期

页      面：695-708页

核心收录：

学科分类：12[管理学] 1201[管理学-管理科学与工程(可授管理学、工学学位)] 081104[工学-模式识别与智能系统] 081203[工学-计算机应用技术] 08[工学] 0835[工学-软件工程] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Key-Area Research and Development Program of Guangdong Province,Grant/Award Number:2021B0101200001 National Natural Science Foundation of China,Grant/Award Numbers:61876140,U20B2065,U21B2048 Open Research Projects of Zhejiang Lab,Grant/Award Number:2019KD0AD01/010 

主　　题：artificial intelligence computer vision deep learning 

摘      要：Segmenting the semantic regions of point clouds is a crucial step for intelligent agents to understand 3D *** supervised point cloud segmentation is highly desirable because entirely labelling point clouds is highly time-consuming and *** the low-costing labelling of 3D point clouds,the scene-level label is one of the most effortless label ***,due to the limitation of classifier discriminative capability and the orderless and structurless nature of the point cloud data,existing scene-level method is hard to transfer the semantic information,which usually leads to the under-activated or over-activated *** this end,a local semantic embedding network is introduced to learn local structural patterns and semantic ***,the proposed network contains graph convolution-based dilation and erosion embedding modules to implement‘inside-out’and‘outside-in’semantic information dissemination ***,the proposed weakly supervised learning framework could achieve the mutual propagation of semantic information in the foreground and *** experiments on the widely used ScanNet benchmark demonstrate the superior capacity of the proposed approach when compared to the current alternatives and baseline models.