Article(id=1329420235100238821, tenantId=1146029695717560320, journalId=1324759914146529478, issueId=1329420206000157303, articleNumber=null, orderNo=null, doi=10.7544/issn1000-1239.202550794, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1762099200000, receivedDateStr=2025-11-03, revisedDate=1776182400000, revisedDateStr=2026-04-15, acceptedDate=1777478400000, acceptedDateStr=2026-04-30, onlineDate=1794888083480, onlineDateStr=2026-11-17, pubDate=1785513600000, pubDateStr=2026-08-01, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1794888083480, onlineIssueDateStr=2026-11-17, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1794888083480, creator=13701087609, updateTime=1794888083480, updator=13701087609, issue=Issue{id=1329420206000157303, tenantId=1146029695717560320, journalId=1324759914146529478, year='2026', volume='63', issue='8', pageStart='1905', pageEnd='2181', issueExtLink='null', onlineDate='null', pubDate='1785513600000', pubDateStr='2026-08-01', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1794888076542, creator='13701087609', updateTime=1794888076542, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext=null, issueFiles=null, downloadFileDto=null}, startPage=2154, endPage=2168, ext={EN=ArticleExt(id=1329420235330925542, articleId=1329420235100238821, tenantId=1146029695717560320, journalId=1324759914146529478, language=EN, title=A Geometric Point Cloud Completion Network Based on Hypergraph Convolution, columnId=1325489891789910943, journalTitle=Journal of Computer Research and Development, columnName=Artificial Intelligence, runingTitle=null, highlight=null, articleAbstract=

Deep learning methods have achieved remarkable progress in point cloud completion tasks due to their fast inference and good generalization capabilities. However, existing methods pay less attention to the local geometric information of point clouds, such as normal vectors, vector angles, and higher-order relationships. To address this issue, we propose a geometric point cloud completion network based on hypergraph convolution to generate detailed and complete point clouds. The framework mainly consists of two core components: a geometric feature extractor and a coarse generation module based on hypergraph convolution. Specifically, the former fully extracts the normal vectors, the difference vectors between the center points and the sampled points, and their angles, and integrates them as position encodings into the Transformer structure, enabling the encoder to more fully learn the geometric characteristics and rotation-invariant features of the point cloud. Additionally, in the coarse generation module based on hypergraph convolution, by introducing hypergraph modeling to capture the higher-order correlations between points and using the cross-attention mechanism to explore the interaction between per-point features and higher-order neighborhood features, it guides the generation of point clouds with more geometric structural rationality. Extensive experimental results on multiple datasets verify that the proposed method has superior performance.

, authors=Jiayue Qiu1, Hailiang Ye1, Jiye Liang2, Ke Lü3, Feilong Cao4, *, authorsList=Jiayue Qiu, Hailiang Ye, Jiye Liang, Ke Lü, Feilong Cao, authorCompany=null, correspAuthors=Feilong Cao, authorNote=

born in 2001. Master candidate. Her main research interests include deep learning and point cloud analysis

born in 1990. PhD, associate professor, master supervisor. Member of CCF. His main research interests include deep learning and point cloud analysis

born in 1962. PhD, professor, PhD supervisor. Fellow of CCF. His main research interests include artificial intelligence, granular computing, and data mining

born in 1971. PhD, professor, PhD supervisor. Member of CCF. His main research interests include computer vision and computer graphics

born in 1965. PhD, professor, PhD supervisor. His main research interests include deep learning and point cloud analysis

, correspAuthorsNote=null, copyrightStatement=Copyright © 2026 Journal of Computer Research and Development. All rights reserved., copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, fund=null), CN=ArticleExt(id=1329420237759427583, articleId=1329420235100238821, tenantId=1146029695717560320, journalId=1324759914146529478, language=CN, title=基于超图卷积的几何点云补全网络, columnId=1325489892104483745, journalTitle=计算机研究与发展, columnName=人工智能, runingTitle=null, highlight=null, articleAbstract=

深度学习方法因其具备快速推理和良好泛化能力,在点云补全任务中取得了显著进展。然而,现有方法较少关注点云的局部几何信息,如法线、向量夹角及高阶关系等。为此,提出一种基于超图卷积的几何点云补全网络,以生成细节丰富的完整点云。该框架主要包括2个核心组件:几何特征提取器与基于超图卷积的粗生成模块。具体而言,前者充分提取点云的法线、中心点与采样点之间的差异向量及其夹角,并将其作为位置编码融入Transformer结构中,使编码器更充分地学习点云的几何特性与旋转不变特征。此外,在基于超图卷积的粗生成模块中,通过引入超图建模点之间的高阶相关性,并利用交叉注意力机制探索逐点特征与高阶邻域特征间的交互作用,指导生成更具几何结构合理性的点云。在多个数据集上的大量实验结果验证了该方法具有较为优越的性能。

, authors=邱佳悦1, 叶海良1, 梁吉业2, 吕科3, 曹飞龙4, *, authorsList=邱佳悦, 叶海良, 梁吉业, 吕科, 曹飞龙, authorCompany=null, correspAuthors=曹飞龙, authorNote=

2001年生。硕士研究生。主要研究方向为深度学习、点云分析

1990年生。博士,副教授,硕士生导师。CCF会员。主要研究方向为深度学习、点云分析

1962年生。博士,教授,博士生导师。CCF会士。主要研究方向为人工智能、粒计算、数据挖掘

1971年生。博士,教授,博士生导师。CCF会员。主要研究方向为计算机视觉、计算机图形学

1965年生。博士,教授,博士生导师。主要研究方向为深度学习、点云分析

, correspAuthorsNote=
曹飞龙()
, copyrightStatement=版权所有 © 2026 计算机研究与发展编辑部, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=Pa+zzPD43oGa5k/WYTzZVw==, magXml=EZmiaY7+OwVLIn5cpn62Wg==, pdfUrl=null, pdf=NONRBAz8m0qTzJi3ACmy3A==, pdfFileSize=3444562, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=3t3BrQagOTT66jIdDWFfIQ==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=ewLMvy+b9xR1oRv9l6HLww==, mapNumber=null, fund=null)}, authors=[Author(id=1329420238338240524, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=qiujy0308@cjlu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420238443098126, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238338240524, language=EN, stringName=Jiayue Qiu, firstName=Jiayue, middleName=null, lastName=Qiu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1College of Sciences, China Jiliang University, Hangzhou 310018, bio={"content":"

born in 2001. Master candidate. Her main research interests include deep learning and point cloud analysis

"}, bioImg=null, bioContent=

born in 2001. Master candidate. Her main research interests include deep learning and point cloud analysis

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420238514401295, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238338240524, language=CN, stringName=邱佳悦, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1中国计量大学理学院 杭州 310018, bio={"img":"qbTTr4IwmC1DsxDKtnDdcw==","content":"

2001年生。硕士研究生。主要研究方向为深度学习、点云分析

"}, bioImg=qbTTr4IwmC1DsxDKtnDdcw==, bioContent=

2001年生。硕士研究生。主要研究方向为深度学习、点云分析

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420237994308608, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=1, ext=[AuthorCompanyExt(id=1329420238002696192, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Sciences, China Jiliang University, Hangzhou 310018), AuthorCompanyExt(id=1329420238011084801, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1中国计量大学理学院 杭州 310018)])]), Author(id=1329420238589898769, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420238669590547, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238589898769, language=EN, stringName=Hailiang Ye, firstName=Hailiang, middleName=null, lastName=Ye, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1College of Sciences, China Jiliang University, Hangzhou 310018, bio={"content":"

born in 1990. PhD, associate professor, master supervisor. Member of CCF. His main research interests include deep learning and point cloud analysis

"}, bioImg=null, bioContent=

born in 1990. PhD, associate professor, master supervisor. Member of CCF. His main research interests include deep learning and point cloud analysis

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420238728310804, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238589898769, language=CN, stringName=叶海良, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1中国计量大学理学院 杭州 310018, bio={"img":"0faZ9X9r2C8kvG64YfGSwA==","content":"

1990年生。博士,副教授,硕士生导师。CCF会员。主要研究方向为深度学习、点云分析

"}, bioImg=0faZ9X9r2C8kvG64YfGSwA==, bioContent=

1990年生。博士,副教授,硕士生导师。CCF会员。主要研究方向为深度学习、点云分析

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420237994308608, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=1, ext=[AuthorCompanyExt(id=1329420238002696192, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Sciences, China Jiliang University, Hangzhou 310018), AuthorCompanyExt(id=1329420238011084801, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1中国计量大学理学院 杭州 310018)])]), Author(id=1329420238791225366, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420238875111448, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238791225366, language=EN, stringName=Jiye Liang, firstName=Jiye, middleName=null, lastName=Liang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Key Laboratory of Computational Intelligence and Chinese Information Processing (Shanxi University), Ministry of Education, Taiyuan030006, bio={"content":"

born in 1962. PhD, professor, PhD supervisor. Fellow of CCF. His main research interests include artificial intelligence, granular computing, and data mining

"}, bioImg=null, bioContent=

born in 1962. PhD, professor, PhD supervisor. Fellow of CCF. His main research interests include artificial intelligence, granular computing, and data mining

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420238958997529, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238791225366, language=CN, stringName=梁吉业, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2计算智能与中文信息处理教育部重点实验室(山西大学) 太原 030006, bio={"img":"tA0nKQa7y1Td+XXyEd5GXw==","content":"

1962年生。博士,教授,博士生导师。CCF会士。主要研究方向为人工智能、粒计算、数据挖掘

"}, bioImg=tA0nKQa7y1Td+XXyEd5GXw==, bioContent=

1962年生。博士,教授,博士生导师。CCF会士。主要研究方向为人工智能、粒计算、数据挖掘

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420238069805058, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=2, ext=[AuthorCompanyExt(id=1329420238078193667, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238069805058, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Key Laboratory of Computational Intelligence and Chinese Information Processing (Shanxi University), Ministry of Education, Taiyuan030006), AuthorCompanyExt(id=1329420238082387972, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238069805058, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2计算智能与中文信息处理教育部重点实验室(山西大学) 太原 030006)])]), Author(id=1329420239030300699, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420239105798173, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420239030300699, language=EN, stringName=Ke Lü, firstName=Ke, middleName=null, lastName=Lü, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3School of Engineering Science, University of Chinese Academy of Sciences, Beijing 101408, bio={"content":"

born in 1971. PhD, professor, PhD supervisor. Member of CCF. His main research interests include computer vision and computer graphics

"}, bioImg=null, bioContent=

born in 1971. PhD, professor, PhD supervisor. Member of CCF. His main research interests include computer vision and computer graphics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420239219044382, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420239030300699, language=CN, stringName=吕科, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3中国科学院大学工程科学学院 北京 101408, bio={"img":"bT3o1S0xRm0v89r8m9d1Wg==","content":"

1971年生。博士,教授,博士生导师。CCF会员。主要研究方向为计算机视觉、计算机图形学

"}, bioImg=bT3o1S0xRm0v89r8m9d1Wg==, bioContent=

1971年生。博士,教授,博士生导师。CCF会员。主要研究方向为计算机视觉、计算机图形学

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420238162079749, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=3, ext=[AuthorCompanyExt(id=1329420238170468358, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238162079749, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3School of Engineering Science, University of Chinese Academy of Sciences, Beijing 101408), AuthorCompanyExt(id=1329420238178856967, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238162079749, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3中国科学院大学工程科学学院 北京 101408)])]), Author(id=1329420239286153248, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=caofeilong88@zjnu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1329420239361650722, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420239286153248, language=EN, stringName=Feilong Cao, firstName=Feilong, middleName=null, lastName=Cao, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=4, *, address=4School of Mathematical Sciences, Zhejiang Normal University, Jinhua, Zhejiang 321014, bio={"content":"

born in 1965. PhD, professor, PhD supervisor. His main research interests include deep learning and point cloud analysis

"}, bioImg=null, bioContent=

born in 1965. PhD, professor, PhD supervisor. His main research interests include deep learning and point cloud analysis

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420239428759587, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420239286153248, language=CN, stringName=曹飞龙, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=4, *, address=4浙江师范大学数学科学学院 浙江金华 321014, bio={"img":"1/mEn+lIJLGJwNIYPV7cgQ==","content":"

1965年生。博士,教授,博士生导师。主要研究方向为深度学习、点云分析

"}, bioImg=1/mEn+lIJLGJwNIYPV7cgQ==, bioContent=

1965年生。博士,教授,博士生导师。主要研究方向为深度学习、点云分析

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420238258548744, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=4, ext=[AuthorCompanyExt(id=1329420238266937353, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238258548744, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=4School of Mathematical Sciences, Zhejiang Normal University, Jinhua, Zhejiang 321014), AuthorCompanyExt(id=1329420238271131658, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238258548744, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=4浙江师范大学数学科学学院 浙江金华 321014)])])], keywords=[Keyword(id=1329420239525228580, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, orderNo=1, keyword=graph neural network), Keyword(id=1329420239604920357, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, orderNo=2, keyword=point cloud completion), Keyword(id=1329420239663640614, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, orderNo=3, keyword=geometric deep learning), Keyword(id=1329420239726555175, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, orderNo=4, keyword=hypergraph learning), Keyword(id=1329420239793664040, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, orderNo=5, keyword=feature extraction), Keyword(id=1329420239856578601, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, orderNo=1, keyword=图神经网络), Keyword(id=1329420239923687466, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, orderNo=2, keyword=点云补全), Keyword(id=1329420239994990635, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, orderNo=3, keyword=几何深度学习), Keyword(id=1329420240053710892, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, orderNo=4, keyword=超图学习), Keyword(id=1329420240129208365, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, orderNo=5, keyword=特征提取)], refs=[Reference(id=1329420243165884491, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=null, journalName=null, refType=null, unstructuredReference=Song Ziying, Liu Lin, Jia Feiyang, et al. Robustness-aware 3D object detection in autonomous driving: A review and outlook[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(11): 15407−15436, articleTitle=null, refAbstract=null), Reference(id=1329420243270742092, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=1, authorNames=null, journalName=null, refType=null, unstructuredReference=廖瑞杰, 杨绍发, 孟文霞, 等. SegGraph: 室外场景三维点云闭环检测算法[J]. 计算机研究与发展, 2019, 56(2): 338−348, articleTitle=null, refAbstract=null), Reference(id=1329420243367211085, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=2, authorNames=null, journalName=null, refType=null, unstructuredReference=Liao Ruijie, Yang Shaofa, Meng Wenxia, et al. SegGraph: An algorithm for loop-closure detection in outdoor scenes using 3D point clouds[J]. Journal of Computer Research and Development, 2019, 56(2): 338−348 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329420243459485774, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=3, rfOrder=3, authorNames=null, journalName=null, refType=null, unstructuredReference=Sulzer R, Marlet R, Vallet B, et al. A survey and benchmark of automatic surface reconstruction from point clouds[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(3): 2000−2019, articleTitle=null, refAbstract=null), Reference(id=1329420243551760463, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=4, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang Junming, Chen Weijia, Wang Yuping, et al. Point set voting for partial point cloud analysis[J]. IEEE Robotics and Automation Letters, 2021, 6(2): 596−603, articleTitle=null, refAbstract=null), Reference(id=1329420243673395280, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=5, rfOrder=5, authorNames=null, journalName=null, refType=null, unstructuredReference=Sarkar K, Varanasi K, Stricker D. Learning quadrangulated patches for 3D shape parameterization and completion[C]//Proc of the 5th Int Conf on 3D Vision. Piscataway, NJ: IEEE, 2017: 383−392, articleTitle=null, refAbstract=null), Reference(id=1329420243786641489, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=6, rfOrder=6, authorNames=null, journalName=null, refType=null, unstructuredReference=Qi C R, Su Hao, Mo Kaichun, et al. PointNet: Deep learning on point sets for 3D classification and segmentation[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2017: 77−85, articleTitle=null, refAbstract=null), Reference(id=1329420243899887698, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=7, rfOrder=7, authorNames=null, journalName=null, refType=null, unstructuredReference=Qi C R, Yi Li, Su Hao, et al. PointNet++: Deep hierarchical feature learning on point sets in a metric space[C]//Advances in Neural Information Processing Systems. Cambridge, MA: MIT, 2017: 5105−5114, articleTitle=null, refAbstract=null), Reference(id=1329420244008939603, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=8, rfOrder=8, authorNames=null, journalName=null, refType=null, unstructuredReference=Yuan Wentao, Khot T, Held D, et al. PCN: Point completion network[C]//Proc of the 6th Int Conf on 3D Vision. Piscataway, NJ: IEEE, 2018: 728−737, articleTitle=null, refAbstract=null), Reference(id=1329420244126380116, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=9, rfOrder=9, authorNames=null, journalName=null, refType=null, unstructuredReference=Huang Zitian, Yu Yikuan, Xu Jiawen, et al. PF-Net: Point fractal network for 3D Point cloud completion[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 7659−7667, articleTitle=null, refAbstract=null), Reference(id=1329420244248014933, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=10, rfOrder=10, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang Xiaogang, Ang M H, Lee G H. Cascaded refinement network for point cloud completion[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 787−796, articleTitle=null, refAbstract=null), Reference(id=1329420244323512406, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=11, rfOrder=11, authorNames=null, journalName=null, refType=null, unstructuredReference=Wen Xin, Li Tianyang, Han Zhizhong, et al. Point cloud completion by skip-attention network with hierarchical folding[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 1936−1945, articleTitle=null, refAbstract=null), Reference(id=1329420244390621271, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=12, rfOrder=12, authorNames=null, journalName=null, refType=null, unstructuredReference=Zong Daoming, Sun Shiliang, Zhao Jing. ASHF-Net: Adaptive sampling and hierarchical folding network for robust point cloud completion[C]//Proc of the 35th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2021: 3625−3632, articleTitle=null, refAbstract=null), Reference(id=1329420244449341528, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=13, rfOrder=13, authorNames=null, journalName=null, refType=null, unstructuredReference=Xiang Peng, Wen Xin, Liu Yushen, et al. Snowflake point deconvolution for point cloud completion and generation with skip-Transformer[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(5): 6320−6338, articleTitle=null, refAbstract=null), Reference(id=1329420244516450393, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=14, rfOrder=14, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhang Wenxiao, Zhou Huajian, Dong Zhen, et al. Point cloud completion via skeleton-detail Transformer[J]. IEEE Transactions on Visualization and Computer Graphics, 2023, 29(10): 4229−4242, articleTitle=null, refAbstract=null), Reference(id=1329420244579364954, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=15, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang Yue, Sun Yongbin, Liu Ziwei, et al. Dynamic graph CNN for learning on point clouds[J]. ACM Transactions on Graphics, 2019, 38(5): 1−12, articleTitle=null, refAbstract=null), Reference(id=1329420244642279515, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=16, rfOrder=16, authorNames=null, journalName=null, refType=null, unstructuredReference=Yu Xumin, Rao Yongming, Wang Ziyi, et al. PoinTr: Diverse point cloud completion with geometry-aware Transformers[C]//Proc of the 18th IEEE Int Conf on Computer Vision. Piscataway, NJ: IEEE, 2021: 12478−12487, articleTitle=null, refAbstract=null), Reference(id=1329420244705194076, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=17, authorNames=null, journalName=null, refType=null, unstructuredReference=Yu Xumin, Rao Yongming, Wang Ziyi, et al. AdaPoinTr: Diverse point cloud completion with adaptive geometry-aware Transformers[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(12): 14114−14130, articleTitle=null, refAbstract=null), Reference(id=1329420244772302941, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=18, authorNames=null, journalName=null, refType=null, unstructuredReference=Xiao Haihong, Xu Hongbin, Li Yuqiong, et al. Multi-dimensional graph interactional network for progressive point cloud completion[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1−12, articleTitle=null, refAbstract=null), Reference(id=1329420244843606110, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=19, rfOrder=19, authorNames=null, journalName=null, refType=null, unstructuredReference=Huang Minming, Fu Yanggeng, Liu Genggeng, et al. GMP-Net: Graph based missing part patching network for point cloud completion[J]. Engineering Applications of Artificial Intelligence, 2024, 130: 107656, articleTitle=null, refAbstract=null), Reference(id=1329420244906520671, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=20, rfOrder=20, authorNames=null, journalName=null, refType=null, unstructuredReference=李娇娇, 孙红岩, 董雨, 等. 基于深度学习的3维点云处理综述[J]. 计算机研究与发展, 2022, 59(5): 1160−1179, articleTitle=null, refAbstract=null), Reference(id=1329420244969435232, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=20, rfOrder=21, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Jiaojiao, Sun Hongyan, Dong Yu, et al. Survey of 3-dimensional point cloud processing based on deep learning[J]. Journal of Computer Research and Development, 2022, 59(5): 1160−1179 (in Chinese), articleTitle=null, refAbstract=null), Reference(id=1329420245040738401, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=21, rfOrder=22, authorNames=null, journalName=null, refType=null, unstructuredReference=Zhou Haoran, Cao Yun, Chu Wenqing, et al. SeedFormer: Patch seeds based point cloud completion with upsample Transformer[C]//Proc of the 17th European Conf on Computer Vision. Berlin: Springer, 2022: 416−432, articleTitle=null, refAbstract=null), Reference(id=1329420245112041570, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=22, rfOrder=23, authorNames=null, journalName=null, refType=null, unstructuredReference=Wen Xin, Xiang Peng, Han Zhizhong, et al. PMP-Net++: Point cloud completion by Transformer-enhanced multi-step point moving paths[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(1): 852−867, articleTitle=null, refAbstract=null), Reference(id=1329420245191733347, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=23, rfOrder=24, authorNames=null, journalName=null, refType=null, unstructuredReference=Li Shanshan, Gao Pan, Tan Xiaoyang, et al. ProxyFormer: Proxy alignment assisted point cloud completion with missing part sensitive Transformer[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2023: 9466−9475, articleTitle=null, refAbstract=null), Reference(id=1329420245267230820, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=24, rfOrder=25, authorNames=null, journalName=null, refType=null, unstructuredReference=Wang Jun, Cui Ying, Guo Dongyan, et al. PointAttN: You only need attention for point cloud completion[C]//Proc of the 38th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2024: 5472−5480, articleTitle=null, refAbstract=null), Reference(id=1329420245330145381, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=25, rfOrder=26, authorNames=null, journalName=null, refType=null, unstructuredReference=Rong Yi, Zhou Haoran, Yuan Lixin, et al. CRA-PCN: Point cloud completion with intra- and inter-level cross-resolution Transformers[C]//Proc of the 38th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2024: 4676−4685, articleTitle=null, refAbstract=null), Reference(id=1329420245393059942, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=26, rfOrder=27, authorNames=null, journalName=null, refType=null, unstructuredReference=Liang Hao, He Zhaoshui, Wang Xu, et al. HGG-Net: Hierarchical geometry generation network for point cloud completion[J]. IEEE Transactions on Intelligent Transportation Systems, 2025, 26(9): 12999−13010, articleTitle=null, refAbstract=null), Reference(id=1329420245460168807, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=27, rfOrder=28, authorNames=null, journalName=null, refType=null, unstructuredReference=Fei Ben, Liu Liwen, Luo Tianyue, et al. Point patches contrastive learning for enhanced point cloud completion[J]. IEEE Transactions on Multimedia, 2025, 27: 581−596, articleTitle=null, refAbstract=null), Reference(id=1329420245527277672, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=28, rfOrder=29, authorNames=null, journalName=null, refType=null, unstructuredReference=Fang Chenghao, Liang Jianqing, Liang Jiye, et al. Multi-modal point cloud completion with interleaved attention enhanced Transformer[C]//Proc of the 34th Int Joint Conf on Artificial Intelligence. San Francisco, CA: Morgan Kaufmann, 2025: 963−971, articleTitle=null, refAbstract=null), Reference(id=1329420245585997929, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=29, rfOrder=30, authorNames=null, journalName=null, refType=null, unstructuredReference=Pan Liang. ECG: Edge-aware point cloud completion with graph convolution[J]. IEEE Robotics and Automation Letters, 2020, 5(3): 4392−4398, articleTitle=null, refAbstract=null), Reference(id=1329420245653106794, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=30, rfOrder=31, authorNames=null, journalName=null, refType=null, unstructuredReference=Shi Jieqi, Xu Lingyun, Heng Liang, et al. Graph-guided deformation for point cloud completion[J]. IEEE Robotics and Automation Letters, 2021, 6(4): 7081−7088, articleTitle=null, refAbstract=null), Reference(id=1329420245711827051, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=31, rfOrder=32, authorNames=null, journalName=null, refType=null, unstructuredReference=Du Zhenjiang, Liu Zhitao, Wang Guan, et al. CMNet: Cross-modal coarse-to-fine network for point cloud completion based on patches[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2025, 35(9): 9132−9147, articleTitle=null, refAbstract=null), Reference(id=1329420245770547308, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=32, rfOrder=33, authorNames=null, journalName=null, refType=null, unstructuredReference=Shu Zhenyu, Yao Jian, Xin Shiqing. DFG-PCN: Point cloud completion with degree-flexible point graph[J]. IEEE Transactions on Visualization and Computer Graphics, 2025, 31(12): 1−14, articleTitle=null, refAbstract=null), Reference(id=1329420245846044781, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=33, rfOrder=34, authorNames=null, journalName=null, refType=null, unstructuredReference=Yuan Yongzhe, Wu Yue, Fan Xiaolong, et al. EGST: Enhanced geometric structure Transformer for point cloud registration[J]. IEEE Transactions on Visualization and Computer Graphics, 2024, 30(9): 6222−6234, articleTitle=null, refAbstract=null), Reference(id=1329420245917347950, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=34, rfOrder=35, authorNames=null, journalName=null, refType=null, unstructuredReference=Feng Yifan, You Haoxuan, Zhang Zizhao, et al. Hypergraph neural networks[C]//Proc of the 33rd AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2019: 3558−3565, articleTitle=null, refAbstract=null), Reference(id=1329420245997039727, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=35, rfOrder=36, authorNames=null, journalName=null, refType=null, unstructuredReference=Wu Zhirong, Song Shuran, Khosla A, et al. 3D ShapeNets: A deep representation for volumetric shapes[C]//Proc of IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2015: 1912−1920, articleTitle=null, refAbstract=null), Reference(id=1329420246072537200, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=36, rfOrder=37, authorNames=null, journalName=null, refType=null, unstructuredReference=Wu Weichao, Xu Yongyang, Zeng Ziyin, et al. Dual-attention fusion point cloud completion with spatially aware clustering[J]. Expert Systems with Applications, 2025, 280: 127557, articleTitle=null, refAbstract=null), Reference(id=1329420246143840369, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=37, rfOrder=38, authorNames=null, journalName=null, refType=null, unstructuredReference=Fei Ben, Yang Weidong, Chen Wenming, et al. Comprehensive review of deep learning-based 3D point cloud completion processing and analysis[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(12): 22862−22883, articleTitle=null, refAbstract=null), Reference(id=1329420246202560626, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=38, rfOrder=39, authorNames=null, journalName=null, refType=null, unstructuredReference=Yang Yaoqing, Feng Chen, Shen Yiru, et al. FoldingNet: Point cloud auto-encoder via deep grid deformation[C]//Proc of IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2018: 206−215, articleTitle=null, refAbstract=null), Reference(id=1329420246265475187, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=39, rfOrder=40, authorNames=null, journalName=null, refType=null, unstructuredReference=Tchapmi L P, Kosaraju V, Rezatofighi H, et al. TopNet: Structural point cloud decoder[C]//Proc of IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2019: 383−392, articleTitle=null, refAbstract=null), Reference(id=1329420246320001140, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=40, rfOrder=41, authorNames=null, journalName=null, refType=null, unstructuredReference=Liu Minghua, Sheng Lu, Yang Sheng, et al. Morphing and sampling network for dense point cloud completion[C]//Proc of the 34th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2020: 11596−11603, articleTitle=null, refAbstract=null), Reference(id=1329420246395498613, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=41, rfOrder=42, authorNames=null, journalName=null, refType=null, unstructuredReference=Cai Pingping, Scott D, Li Xiaoguang, et al. Orthogonal dictionary guided shape completion network for point cloud[C]//Proc of the 38th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2024: 864−872, articleTitle=null, refAbstract=null), Reference(id=1329420246462607478, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=42, rfOrder=43, authorNames=null, journalName=null, refType=null, unstructuredReference=Cai Pingping, Zhang Canyu, Shi Lingjia, et al. EINet: Point cloud completion via extrapolation and interpolation[C]//Proc of the 18th European Conf on Computer Vision. Berlin: Springer, 2025: 377−393, articleTitle=null, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1329420237994308608, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=1, ext=[AuthorCompanyExt(id=1329420238002696192, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Sciences, China Jiliang University, Hangzhou 310018), AuthorCompanyExt(id=1329420238011084801, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1中国计量大学理学院 杭州 310018)]), AuthorCompany(id=1329420238069805058, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=2, ext=[AuthorCompanyExt(id=1329420238078193667, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238069805058, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Key Laboratory of Computational Intelligence and Chinese Information Processing (Shanxi University), Ministry of Education, Taiyuan030006), AuthorCompanyExt(id=1329420238082387972, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238069805058, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2计算智能与中文信息处理教育部重点实验室(山西大学) 太原 030006)]), AuthorCompany(id=1329420238162079749, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=3, ext=[AuthorCompanyExt(id=1329420238170468358, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238162079749, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3School of Engineering Science, University of Chinese Academy of Sciences, Beijing 101408), AuthorCompanyExt(id=1329420238178856967, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238162079749, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3中国科学院大学工程科学学院 北京 101408)]), AuthorCompany(id=1329420238258548744, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=4, ext=[AuthorCompanyExt(id=1329420238266937353, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238258548744, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=4School of Mathematical Sciences, Zhejiang Normal University, Jinhua, Zhejiang 321014), AuthorCompanyExt(id=1329420238271131658, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238258548744, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=4浙江师范大学数学科学学院 浙江金华 321014)])], figs=[ArticleFig(id=1329420240271814702, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Figure 1, caption=Overall schematic diagram of the geometric point cloud completion network based on hypergraph convolution, figureFileSmall=5EJdoBm5HRIpb1FlR5tAYQ==, figureFileBig=3t3BrQagOTT66jIdDWFfIQ==, tableContent=null), ArticleFig(id=1329420240351506479, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=图1, caption=基于超图卷积的几何点云补全网络整体示意图, figureFileSmall=5EJdoBm5HRIpb1FlR5tAYQ==, figureFileBig=3t3BrQagOTT66jIdDWFfIQ==, tableContent=null), ArticleFig(id=1329420240557027376, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Figure 2, caption=Details of the geometric Transformer, figureFileSmall=CUYZjh2BvHSKQlz3VexyGQ==, figureFileBig=2+n2YjTLL81vWrFRH0H18g==, tableContent=null), ArticleFig(id=1329420240640913457, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=图2, caption=几何Transformer的具体细节, figureFileSmall=CUYZjh2BvHSKQlz3VexyGQ==, figureFileBig=2+n2YjTLL81vWrFRH0H18g==, tableContent=null), ArticleFig(id=1329420240712216626, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Figure 3, caption=Geometric feature extractor, figureFileSmall=37TMEh/J7+wEkPGx0HmDmQ==, figureFileBig=xYYhVk9gbnvt53QFk0TXGA==, tableContent=null), ArticleFig(id=1329420240783519795, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=图3, caption=几何特征提取器, figureFileSmall=37TMEh/J7+wEkPGx0HmDmQ==, figureFileBig=xYYhVk9gbnvt53QFk0TXGA==, tableContent=null), ArticleFig(id=1329420240850628660, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Figure 4, caption=Coarse generation module based on hypergraph convolution, figureFileSmall=SMZMnnaPAdFC972SmaJhvA==, figureFileBig=Y2TjXrLvoahFS3/Y8/7Mbw==, tableContent=null), ArticleFig(id=1329420240917737525, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=图4, caption=基于超图卷积的粗生成模块, figureFileSmall=SMZMnnaPAdFC972SmaJhvA==, figureFileBig=Y2TjXrLvoahFS3/Y8/7Mbw==, tableContent=null), ArticleFig(id=1329420240984846390, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Figure 5, caption=Visualization results of various methods on PCN dataset, figureFileSmall=kb0B+kgnXOwhSwXCHbTA1w==, figureFileBig=UaXhV6EHpMCBxrhs+krWlw==, tableContent=null), ArticleFig(id=1329420241064538167, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=图5, caption=在PCN数据集上各类方法的可视化结果, figureFileSmall=kb0B+kgnXOwhSwXCHbTA1w==, figureFileBig=UaXhV6EHpMCBxrhs+krWlw==, tableContent=null), ArticleFig(id=1329420241144229945, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Figure 6, caption=Category comparison, figureFileSmall=nOTg1khCameBGvWTSvNeBg==, figureFileBig=vEIGHXhzCcBnPCtxfwVnPg==, tableContent=null), ArticleFig(id=1329420241207144506, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=图6, caption=类别对比, figureFileSmall=nOTg1khCameBGvWTSvNeBg==, figureFileBig=vEIGHXhzCcBnPCtxfwVnPg==, tableContent=null), ArticleFig(id=1329420241278447675, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Figure 7, caption=Visual comparison of ablation experiments, figureFileSmall=+BhFi5u4hnwgO85KXHUYEQ==, figureFileBig=6AXqVUuNet+3f7ZeQirkFg==, tableContent=null), ArticleFig(id=1329420241374916668, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=图7, caption=消融实验的可视化对比, figureFileSmall=+BhFi5u4hnwgO85KXHUYEQ==, figureFileBig=6AXqVUuNet+3f7ZeQirkFg==, tableContent=null), ArticleFig(id=1329420241458802749, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Table 1, caption=

Detailed Configuration of the Proposed Network Framework

, figureFileSmall=null, figureFileBig=null, tableContent=
步骤阶段输入维度输出维度特征维度坐标维度
输入编码器2048 $ \times $3
集抽象层3128512$ \times $128512$ \times $3
几何Transformer128128512$ \times $128
集抽象层128256128$ \times $256128$ \times $3
几何Transformer256256128$ \times $256
超图卷积解码器256256128$ \times $256
交叉注意力256256128$ \times $256
MLP256128256$ \times $128
MLP128128256$ \times $128
MLP128128256$ \times $128
MLP1283256$ \times $3
最远点采样33512$ \times $3
细化模块31282048 $ \times $1282048 $ \times $3
输出2048 $ \times $3
), ArticleFig(id=1329420241559466046, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=表1, caption=

本文网络框架的详细配置

, figureFileSmall=null, figureFileBig=null, tableContent=
步骤阶段输入维度输出维度特征维度坐标维度
输入编码器2048 $ \times $3
集抽象层3128512$ \times $128512$ \times $3
几何Transformer128128512$ \times $128
集抽象层128256128$ \times $256128$ \times $3
几何Transformer256256128$ \times $256
超图卷积解码器256256128$ \times $256
交叉注意力256256128$ \times $256
MLP256128256$ \times $128
MLP128128256$ \times $128
MLP128128256$ \times $128
MLP1283256$ \times $3
最远点采样33512$ \times $3
细化模块31282048 $ \times $1282048 $ \times $3
输出2048 $ \times $3
), ArticleFig(id=1329420241672712255, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Table 2, caption=

CD-T, CD-P, and F-Score Results on PCN Dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
指标方法飞机橱柜汽车椅子灯具沙发桌子船只平均
$ {10}^{4}\times CD\text{-}T \downarrow $PCN[8]2.799.515.869.499.4111.387.976.287.84
FoldingNet[38]3.8110.695.9512.2811.8812.0211.467.619.46
TopNet[39]5.4715.038.0714.9715.8115.4912.449.1912.06
MSN[40]2.8312.756.939.338.0213.487.515.818.33
ECG[29]3.3910.747.238.749.6311.059.377.308.43
SeedFormer[21]1.858.575.726.114.269.165.063.565.54
SnowflakeNet[13]2.1210.105.916.785.039.025.683.976.08
SDT[14]2.299.425.518.036.219.486.574.446.49
PointAttN[24]2.029.095.526.595.098.775.674.075.85
CRA-PCN[25]1.798.145.325.463.687.204.653.474.96
ODGNet[41]1.618.185.345.743.988.424.613.365.14
EINet[42]2.028.335.435.844.328.725.013.615.41
本文1.577.895.105.253.107.213.873.244.65
${10}^{3}\times CD\text{-}P \downarrow $PCN[8]8.95318.06814.83417.62016.44218.86115.27714.38015.512
FoldingNet[38]10.75019.22015.02620.02219.54620.32517.79415.83317.315
TopNet[39]12.73323.20117.37921.94422.41822.42819.44917.12619.585
MSN[40]8.99319.35316.12817.04213.71019.37514.96413.58815.394
ECG[29]9.79718.76116.25116.59615.96419.48115.77414.16715.849
SeedFormer[21]7.70116.47614.16313.49210.02316.60011.91510.63412.626
SnowflakeNet[13]7.87017.29514.80214.36711.10716.70312.85311.28113.285
SDT[14]8.04317.76114.36715.71612.28617.69113.75411.73613.919
PointAttN[24]7.64517.08014.31014.29111.18316.50212.89811.13913.131
CRA-PCN[25]7.22316.23213.96313.0979.42615.39211.78910.49112.202
ODGNet[41]6.97816.11213.92013.1499.56115.59011.54710.26312.141
EINet[42]7.31216.57514.39413.5369.95816.17112.12810.49912.572
本文6.84515.95813.69912.7148.95715.15311.15410.10011.823
$ F\text{-}S core\uparrow $PCN[8]0.7370.1980.2660.2480.3570.1910.3360.4100.342
FoldingNet[38]0.5930.1490.2550.1670.1840.1400.2960.2910.259
TopNet[39]0.5110.1200.2030.1850.2150.1410.2460.3050.241
MSN[40]0.7230.2020.2510.2980.5220.2050.3450.4510.375
ECG[29]0.6840.2090.2510.3140.4310.1890.3560.4380.359
SeedFormer[21]0.8180.2660.3070.4310.6770.3000.4600.5930.482
SnowflakeNet[13]0.7870.2600.2900.3960.6240.2830.4370.5400.452
SDT[14]0.7790.2350.2970.3430.5730.2420.3970.5220.424
PointAttN[24]0.8030.2580.3090.3910.6210.2810.4300.5640.457
CRA-PCN[25]0.8300.2790.3250.4420.7100.3150.4910.5980.499
ODGNet[41]0.8430.2880.3390.4490.7070.3260.5080.6160.509
EINet[42]0.8020.2780.3220.4400.6760.3070.4770.5940.487
本文0.8550.2870.3370.4610.7260.3260.5150.6250.517
), ArticleFig(id=1329420241811124288, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=表2, caption=

在PCN数据集上CD-T,CD-P,F-Score结果

, figureFileSmall=null, figureFileBig=null, tableContent=
指标方法飞机橱柜汽车椅子灯具沙发桌子船只平均
$ {10}^{4}\times CD\text{-}T \downarrow $PCN[8]2.799.515.869.499.4111.387.976.287.84
FoldingNet[38]3.8110.695.9512.2811.8812.0211.467.619.46
TopNet[39]5.4715.038.0714.9715.8115.4912.449.1912.06
MSN[40]2.8312.756.939.338.0213.487.515.818.33
ECG[29]3.3910.747.238.749.6311.059.377.308.43
SeedFormer[21]1.858.575.726.114.269.165.063.565.54
SnowflakeNet[13]2.1210.105.916.785.039.025.683.976.08
SDT[14]2.299.425.518.036.219.486.574.446.49
PointAttN[24]2.029.095.526.595.098.775.674.075.85
CRA-PCN[25]1.798.145.325.463.687.204.653.474.96
ODGNet[41]1.618.185.345.743.988.424.613.365.14
EINet[42]2.028.335.435.844.328.725.013.615.41
本文1.577.895.105.253.107.213.873.244.65
${10}^{3}\times CD\text{-}P \downarrow $PCN[8]8.95318.06814.83417.62016.44218.86115.27714.38015.512
FoldingNet[38]10.75019.22015.02620.02219.54620.32517.79415.83317.315
TopNet[39]12.73323.20117.37921.94422.41822.42819.44917.12619.585
MSN[40]8.99319.35316.12817.04213.71019.37514.96413.58815.394
ECG[29]9.79718.76116.25116.59615.96419.48115.77414.16715.849
SeedFormer[21]7.70116.47614.16313.49210.02316.60011.91510.63412.626
SnowflakeNet[13]7.87017.29514.80214.36711.10716.70312.85311.28113.285
SDT[14]8.04317.76114.36715.71612.28617.69113.75411.73613.919
PointAttN[24]7.64517.08014.31014.29111.18316.50212.89811.13913.131
CRA-PCN[25]7.22316.23213.96313.0979.42615.39211.78910.49112.202
ODGNet[41]6.97816.11213.92013.1499.56115.59011.54710.26312.141
EINet[42]7.31216.57514.39413.5369.95816.17112.12810.49912.572
本文6.84515.95813.69912.7148.95715.15311.15410.10011.823
$ F\text{-}S core\uparrow $PCN[8]0.7370.1980.2660.2480.3570.1910.3360.4100.342
FoldingNet[38]0.5930.1490.2550.1670.1840.1400.2960.2910.259
TopNet[39]0.5110.1200.2030.1850.2150.1410.2460.3050.241
MSN[40]0.7230.2020.2510.2980.5220.2050.3450.4510.375
ECG[29]0.6840.2090.2510.3140.4310.1890.3560.4380.359
SeedFormer[21]0.8180.2660.3070.4310.6770.3000.4600.5930.482
SnowflakeNet[13]0.7870.2600.2900.3960.6240.2830.4370.5400.452
SDT[14]0.7790.2350.2970.3430.5730.2420.3970.5220.424
PointAttN[24]0.8030.2580.3090.3910.6210.2810.4300.5640.457
CRA-PCN[25]0.8300.2790.3250.4420.7100.3150.4910.5980.499
ODGNet[41]0.8430.2880.3390.4490.7070.3260.5080.6160.509
EINet[42]0.8020.2780.3220.4400.6760.3070.4770.5940.487
本文0.8550.2870.3370.4610.7260.3260.5150.6250.517
), ArticleFig(id=1329420241957924929, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Table 3, caption=

CD-T, CD-P, and F-Score Results on ShapeNet Extension Dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
指标方法床长凳书架公交车吉他摩托车手枪滑板平均
${10}^{4}\times CD\text{-}T \downarrow $PCN[8]31.328.657.0114.956.356.919.4711.5312.02
ECG[29]25.917.485.9213.095.785.697.3613.4010.58
SnowflakeNet[13]23.976.815.7410.685.453.525.8910.679.09
SDT[14]27.777.326.5813.905.605.856.999.8110.48
PointAttN[24]27.717.275.9612.275.292.706.336.569.26
CRA-PCN[25]20.946.635.3310.194.821.976.156.167.77
ODGNet[41]21.036.494.6110.544.992.845.595.727.76
EINet[42]22.546.904.6511.335.612.445.059.978.56
本文22.436.625.1511.235.472.185.456.718.16
${10}^{3}\times CD\text{-}P \downarrow $PCN[8]26.91014.90313.83519.60313.28914.39317.17117.36917.184
ECG[29]23.07612.95812.31717.07212.43512.21314.45217.97015.312
SnowflakeNet[13]21.23212.41911.06015.42212.0348.38513.05413.18613.349
SDT[14]23.52212.58612.45816.98712.58712.08513.53014.24814.751
PointAttN[24]22.78712.81311.46217.04012.0248.17313.29711.83313.679
CRA-PCN[25]20.79011.99210.66815.38411.6907.95012.40710.68513.273
ODGNet[41]20.44312.0029.99015.44111.5337.70912.48010.60812.525
EINet[42]20.72412.14710.00615.76911.9047.36811.88912.32412.766
本文20.49011.86910.40015.49011.8047.10312.11310.68212.494
$ F\text{-}S core\uparrow $PCN[8]0.1690.4010.4610.2440.4170.4160.2780.3240.339
ECG[29]0.3010.5170.5490.3540.4710.5340.4130.4030.443
SnowflakeNet[13]0.3330.5300.6050.3740.4930.7060.4250.5190.498
SDT[14]0.3090.5290.5610.3570.4540.5540.4500.4680.460
PointAttN[24]0.3010.5120.5960.3280.4840.7050.4250.5380.486
CRA-PCN[25]0.3490.5620.6240.3820.5260.7700.4750.5910.532
ODGNet[41]0.3490.5600.6450.3870.5310.7230.4580.5920.531
EINet[42]0.3270.5350.6220.3610.4990.7240.4640.5490.510
本文0.3510.5720.6400.3880.5270.7520.4790.6010.539
), ArticleFig(id=1329420242083754050, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=表3, caption=

在ShapeNet扩展数据集上CD-T,CD-P,F-Score结果

, figureFileSmall=null, figureFileBig=null, tableContent=
指标方法床长凳书架公交车吉他摩托车手枪滑板平均
${10}^{4}\times CD\text{-}T \downarrow $PCN[8]31.328.657.0114.956.356.919.4711.5312.02
ECG[29]25.917.485.9213.095.785.697.3613.4010.58
SnowflakeNet[13]23.976.815.7410.685.453.525.8910.679.09
SDT[14]27.777.326.5813.905.605.856.999.8110.48
PointAttN[24]27.717.275.9612.275.292.706.336.569.26
CRA-PCN[25]20.946.635.3310.194.821.976.156.167.77
ODGNet[41]21.036.494.6110.544.992.845.595.727.76
EINet[42]22.546.904.6511.335.612.445.059.978.56
本文22.436.625.1511.235.472.185.456.718.16
${10}^{3}\times CD\text{-}P \downarrow $PCN[8]26.91014.90313.83519.60313.28914.39317.17117.36917.184
ECG[29]23.07612.95812.31717.07212.43512.21314.45217.97015.312
SnowflakeNet[13]21.23212.41911.06015.42212.0348.38513.05413.18613.349
SDT[14]23.52212.58612.45816.98712.58712.08513.53014.24814.751
PointAttN[24]22.78712.81311.46217.04012.0248.17313.29711.83313.679
CRA-PCN[25]20.79011.99210.66815.38411.6907.95012.40710.68513.273
ODGNet[41]20.44312.0029.99015.44111.5337.70912.48010.60812.525
EINet[42]20.72412.14710.00615.76911.9047.36811.88912.32412.766
本文20.49011.86910.40015.49011.8047.10312.11310.68212.494
$ F\text{-}S core\uparrow $PCN[8]0.1690.4010.4610.2440.4170.4160.2780.3240.339
ECG[29]0.3010.5170.5490.3540.4710.5340.4130.4030.443
SnowflakeNet[13]0.3330.5300.6050.3740.4930.7060.4250.5190.498
SDT[14]0.3090.5290.5610.3570.4540.5540.4500.4680.460
PointAttN[24]0.3010.5120.5960.3280.4840.7050.4250.5380.486
CRA-PCN[25]0.3490.5620.6240.3820.5260.7700.4750.5910.532
ODGNet[41]0.3490.5600.6450.3870.5310.7230.4580.5920.531
EINet[42]0.3270.5350.6220.3610.4990.7240.4640.5490.510
本文0.3510.5720.6400.3880.5270.7520.4790.6010.539
), ArticleFig(id=1329420242201194563, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Table 4, caption=

Consistency Results on KITTI Dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
方法一致性
PointAttN0.451×10−3
CRA-PCN0.894×10−3
EINet0.758×10−3
本文0.296×10−3
), ArticleFig(id=1329420242331217988, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=表4, caption=

在KITTI数据集上一致性结果

, figureFileSmall=null, figureFileBig=null, tableContent=
方法一致性
PointAttN0.451×10−3
CRA-PCN0.894×10−3
EINet0.758×10−3
本文0.296×10−3
), ArticleFig(id=1329420242440269893, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Table 5, caption=

The Results of Ablation Studies

, figureFileSmall=null, figureFileBig=null, tableContent=
模型CD-TCD-PF-Score推理时间/s训练时间/h
模型15.4612.5840.48785.4219.03
模型25.1512.2520.50288.5324.13
模型35.2812.3450.49792.3929.70
模型45.0812.2620.49893.1131.97
模型5(本文)4.6511.8230.517106.8458.13
), ArticleFig(id=1329420242553516102, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=表5, caption=

消融实验结果

, figureFileSmall=null, figureFileBig=null, tableContent=
模型CD-TCD-PF-Score推理时间/s训练时间/h
模型15.4612.5840.48785.4219.03
模型25.1512.2520.50288.5324.13
模型35.2812.3450.49792.3929.70
模型45.0812.2620.49893.1131.97
模型5(本文)4.6511.8230.517106.8458.13
), ArticleFig(id=1329420242675150919, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Table 6, caption=

Discussion of Hyperparameters

, figureFileSmall=null, figureFileBig=null, tableContent=
$ m $CD-TCD-PF-Score推理时间/s训练时间/h
164.7711.8700.515115.0952.53
324.8111.9390.514114.3056.81
644.6511.8230.517106.8458.13
$ K $CD-TCD-PF-Score推理时间/s训练时间/h
54.6511.8230.517106.8458.13
104.8111.9420.514111.5756.75
154.7811.9010.515119.7655.97
), ArticleFig(id=1329420242788397128, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=表6, caption=

超参数讨论

, figureFileSmall=null, figureFileBig=null, tableContent=
$ m $CD-TCD-PF-Score推理时间/s训练时间/h
164.7711.8700.515115.0952.53
324.8111.9390.514114.3056.81
644.6511.8230.517106.8458.13
$ K $CD-TCD-PF-Score推理时间/s训练时间/h
54.6511.8230.517106.8458.13
104.8111.9420.514111.5756.75
154.7811.9010.515119.7655.97
), ArticleFig(id=1329420242889060425, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=EN, label=Table 7, caption=

Comparison of Computational Complexity

, figureFileSmall=null, figureFileBig=null, tableContent=
方法参数数量运算次数/GFLOPS推理时间/s
PCN[8]6.86×1063.3178.10
MSN[40]29.53×1065.5984.23
ECG[29]14.29×1065.2583.96
SnowflakeNet[13]19.28×1067.3895.25
SDT[14]14.95×1065.5083.84
PointAttN[24]19.00×10611.8287.89
CRA-PCN[25]4.63×1065.68118.63
EINet[42]2.66×10610.6593.77
本文24.66×10617.04106.84
), ArticleFig(id=1329420242998112330, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, language=CN, label=表7, caption=

计算复杂度比较

, figureFileSmall=null, figureFileBig=null, tableContent=
方法参数数量运算次数/GFLOPS推理时间/s
PCN[8]6.86×1063.3178.10
MSN[40]29.53×1065.5984.23
ECG[29]14.29×1065.2583.96
SnowflakeNet[13]19.28×1067.3895.25
SDT[14]14.95×1065.5083.84
PointAttN[24]19.00×10611.8287.89
CRA-PCN[25]4.63×1065.68118.63
EINet[42]2.66×10610.6593.77
本文24.66×10617.04106.84
)], attaches=null, journal=Journal(id=1324759606947315908, delFlag=0, nameCn=计算机研究与发展, nameEn=Journal of Computer Research and Development, nameHistory1=null, nameHistory2=null, issn=1000-1239, eissn=null, cn=11-1777/TP, coden=null, periodic=0, language=CN, oaType=null, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=k2l8zgqwstuxxevFX2qxVA==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Computer Research and Development, journalRemark=null, publicationField=null, createdTime=1793776903141, updatedTime=1793850698612, createdBy=18614031015, updatedBy=13701087609, firstLetterCn=J, firstLetterEn=J, subjectCode=Electronic Information, subjectName=null, subjectCodeEn=Electronic Information, subjectNameEn=null, picCn=k2l8zgqwstuxxevFX2qxVA==, picEn=r6k0Ry+RD+c/VXj5QarmDw==, jcr=null, cjcr=null, exts=[JournalExt(id=1325069127710286688, language=CN, name=计算机研究与发展, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1793850698640, updatedTime=1793850698640, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://crad.ict.ac.cn/journalx/authorLogOn.action, submissionEditorUrl=https://crad.ict.ac.cn/journalx/editorLogOn.action, submissionReviewUrl=https://crad.ict.ac.cn/journalx/expertLogOn.action, submissionCeEditorUrl=https://crad.ict.ac.cn/journalx/editorInChiefLogOn.action, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1325069127777395553, language=EN, name=Journal of Computer Research and Development, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1793850698656, updatedTime=1793850698656, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://crad.ict.ac.cn/journalx/authorLogOn.action, submissionEditorUrl=https://crad.ict.ac.cn/journalx/editorLogOn.action, submissionReviewUrl=https://crad.ict.ac.cn/journalx/expertLogOn.action, submissionCeEditorUrl=https://crad.ict.ac.cn/journalx/editorInChiefLogOn.action, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1324759914146529478, websiteList=[Website(id=1325070300139892771, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1324759914146529478, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN, language=CN, createTime=1793850978169, createBy=18614031015, updateTime=1793851040259, updateBy=18614031015, name=计算机研究与发展-中文, tplId=1146099689490845704, title=计算机研究与发展, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1325070893751349360, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=articleTextType, value=kx, createTime=1793851119697, updateTime=1793851119697, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893730377837, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=banner, value=null, createTime=1793851119692, updateTime=1793851119692, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893772320883, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=grayFlag, value=0, createTime=1793851119702, updateTime=1793851119702, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893721989228, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=logo, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/file/pic?fileId=U27tDsy717YF7XSPA+Vwdw==, createTime=1793851119690, updateTime=1793851119690, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893784903797, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=minRunFlag, value=0, createTime=1793851119705, updateTime=1793851119705, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893742960751, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/file/pic, createTime=1793851119695, updateTime=1793851119695, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893780709492, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=silenceFlag, value=0, createTime=1793851119704, updateTime=1793851119704, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893738766446, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1793851119694, updateTime=1793851119694, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893755543665, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=themeColor, value=null, createTime=1793851119698, updateTime=1793851119698, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070893763932274, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300139892771, code=themeStyle, value=null, createTime=1793851119700, updateTime=1793851119700, creator=18614031015, updator=18614031015)]), Website(id=1325070300580294698, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1324759914146529478, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN, language=EN, createTime=1793850978274, createBy=18614031015, updateTime=1793851036638, updateBy=18614031015, name=计算机研究与发展-英文, tplId=1146101810881728533, title=Journal of Computer Research and Development, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1325070925288321148, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=articleTextType, value=kx, createTime=1793851127216, updateTime=1793851127216, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925263155321, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=banner, value=null, createTime=1793851127210, updateTime=1793851127210, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925309292671, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=grayFlag, value=0, createTime=1793851127221, updateTime=1793851127221, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925254766712, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=logo, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/file/pic?fileId=U27tDsy717YF7XSPA+Vwdw==, createTime=1793851127208, updateTime=1793851127208, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925326069889, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=minRunFlag, value=0, createTime=1793851127225, updateTime=1793851127225, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925279932539, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/file/pic, createTime=1793851127214, updateTime=1793851127214, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925317681280, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=silenceFlag, value=0, createTime=1793851127223, updateTime=1793851127223, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925271543930, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1793851127212, updateTime=1793851127212, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925292515453, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=themeColor, value=null, createTime=1793851127217, updateTime=1793851127217, creator=18614031015, updator=18614031015), WebsiteProps(id=1325070925300904062, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1325070300580294698, code=themeStyle, value=null, createTime=1793851127219, updateTime=1793851127219, creator=18614031015, updator=18614031015)])], journalTitle=计算机研究与发展, weixinUrl=null, journalUrl=https://crad.ict.ac.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Computer Research and Development, journalPhotoCn=k2l8zgqwstuxxevFX2qxVA==, journalPhotoEn=r6k0Ry+RD+c/VXj5QarmDw==, journalFirstLetter=J, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=, provinceCode=null, provinceName=null, collectFlag=false, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202550794, detailUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/10.7544/issn1000-1239.202550794, pdfUrlCn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/PDF/10.7544/issn1000-1239.202550794, pdfUrlEn=https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/PDF/10.7544/issn1000-1239.202550794, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1785513600000, fullTextJson=null, articleText=null, reference=null)
收藏切换
基于超图卷积的几何点云补全网络
邱佳悦 1 , 叶海良 1 , 梁吉业 2 , 吕科 3 , 曹飞龙 4, *
计算机研究与发展 | 人工智能 2026,63(8): 2154-2168
收起
收藏切换
计算机研究与发展 |人工智能 2026 , 63 (8) : 2154 -2168
基于超图卷积的几何点云补全网络
全屏
[Author(id=1329420238338240524, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=qiujy0308@cjlu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420238443098126, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238338240524, language=EN, stringName=Jiayue Qiu, firstName=Jiayue, middleName=null, lastName=Qiu, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1College of Sciences, China Jiliang University, Hangzhou 310018, bio={"content":"

born in 2001. Master candidate. Her main research interests include deep learning and point cloud analysis

"}, bioImg=null, bioContent=

born in 2001. Master candidate. Her main research interests include deep learning and point cloud analysis

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420238514401295, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238338240524, language=CN, stringName=邱佳悦, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1中国计量大学理学院 杭州 310018, bio={"img":"qbTTr4IwmC1DsxDKtnDdcw==","content":"

2001年生。硕士研究生。主要研究方向为深度学习、点云分析

"}, bioImg=qbTTr4IwmC1DsxDKtnDdcw==, bioContent=

2001年生。硕士研究生。主要研究方向为深度学习、点云分析

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420237994308608, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=1, ext=[AuthorCompanyExt(id=1329420238002696192, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Sciences, China Jiliang University, Hangzhou 310018), AuthorCompanyExt(id=1329420238011084801, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1中国计量大学理学院 杭州 310018)])]), Author(id=1329420238589898769, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420238669590547, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238589898769, language=EN, stringName=Hailiang Ye, firstName=Hailiang, middleName=null, lastName=Ye, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1College of Sciences, China Jiliang University, Hangzhou 310018, bio={"content":"

born in 1990. PhD, associate professor, master supervisor. Member of CCF. His main research interests include deep learning and point cloud analysis

"}, bioImg=null, bioContent=

born in 1990. PhD, associate professor, master supervisor. Member of CCF. His main research interests include deep learning and point cloud analysis

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420238728310804, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238589898769, language=CN, stringName=叶海良, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1中国计量大学理学院 杭州 310018, bio={"img":"0faZ9X9r2C8kvG64YfGSwA==","content":"

1990年生。博士,副教授,硕士生导师。CCF会员。主要研究方向为深度学习、点云分析

"}, bioImg=0faZ9X9r2C8kvG64YfGSwA==, bioContent=

1990年生。博士,副教授,硕士生导师。CCF会员。主要研究方向为深度学习、点云分析

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420237994308608, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=1, ext=[AuthorCompanyExt(id=1329420238002696192, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1College of Sciences, China Jiliang University, Hangzhou 310018), AuthorCompanyExt(id=1329420238011084801, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420237994308608, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1中国计量大学理学院 杭州 310018)])]), Author(id=1329420238791225366, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420238875111448, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238791225366, language=EN, stringName=Jiye Liang, firstName=Jiye, middleName=null, lastName=Liang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2Key Laboratory of Computational Intelligence and Chinese Information Processing (Shanxi University), Ministry of Education, Taiyuan030006, bio={"content":"

born in 1962. PhD, professor, PhD supervisor. Fellow of CCF. His main research interests include artificial intelligence, granular computing, and data mining

"}, bioImg=null, bioContent=

born in 1962. PhD, professor, PhD supervisor. Fellow of CCF. His main research interests include artificial intelligence, granular computing, and data mining

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420238958997529, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420238791225366, language=CN, stringName=梁吉业, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, address=2计算智能与中文信息处理教育部重点实验室(山西大学) 太原 030006, bio={"img":"tA0nKQa7y1Td+XXyEd5GXw==","content":"

1962年生。博士,教授,博士生导师。CCF会士。主要研究方向为人工智能、粒计算、数据挖掘

"}, bioImg=tA0nKQa7y1Td+XXyEd5GXw==, bioContent=

1962年生。博士,教授,博士生导师。CCF会士。主要研究方向为人工智能、粒计算、数据挖掘

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420238069805058, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=2, ext=[AuthorCompanyExt(id=1329420238078193667, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238069805058, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2Key Laboratory of Computational Intelligence and Chinese Information Processing (Shanxi University), Ministry of Education, Taiyuan030006), AuthorCompanyExt(id=1329420238082387972, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238069805058, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2计算智能与中文信息处理教育部重点实验室(山西大学) 太原 030006)])]), Author(id=1329420239030300699, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1329420239105798173, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420239030300699, language=EN, stringName=Ke Lü, firstName=Ke, middleName=null, lastName=Lü, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3School of Engineering Science, University of Chinese Academy of Sciences, Beijing 101408, bio={"content":"

born in 1971. PhD, professor, PhD supervisor. Member of CCF. His main research interests include computer vision and computer graphics

"}, bioImg=null, bioContent=

born in 1971. PhD, professor, PhD supervisor. Member of CCF. His main research interests include computer vision and computer graphics

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420239219044382, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420239030300699, language=CN, stringName=吕科, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3中国科学院大学工程科学学院 北京 101408, bio={"img":"bT3o1S0xRm0v89r8m9d1Wg==","content":"

1971年生。博士,教授,博士生导师。CCF会员。主要研究方向为计算机视觉、计算机图形学

"}, bioImg=bT3o1S0xRm0v89r8m9d1Wg==, bioContent=

1971年生。博士,教授,博士生导师。CCF会员。主要研究方向为计算机视觉、计算机图形学

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420238162079749, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=3, ext=[AuthorCompanyExt(id=1329420238170468358, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238162079749, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3School of Engineering Science, University of Chinese Academy of Sciences, Beijing 101408), AuthorCompanyExt(id=1329420238178856967, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238162079749, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3中国科学院大学工程科学学院 北京 101408)])]), Author(id=1329420239286153248, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=caofeilong88@zjnu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1329420239361650722, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420239286153248, language=EN, stringName=Feilong Cao, firstName=Feilong, middleName=null, lastName=Cao, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=4, *, address=4School of Mathematical Sciences, Zhejiang Normal University, Jinhua, Zhejiang 321014, bio={"content":"

born in 1965. PhD, professor, PhD supervisor. His main research interests include deep learning and point cloud analysis

"}, bioImg=null, bioContent=

born in 1965. PhD, professor, PhD supervisor. His main research interests include deep learning and point cloud analysis

, aboutCorrespAuthor=null), CN=AuthorExt(id=1329420239428759587, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, authorId=1329420239286153248, language=CN, stringName=曹飞龙, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=4, *, address=4浙江师范大学数学科学学院 浙江金华 321014, bio={"img":"1/mEn+lIJLGJwNIYPV7cgQ==","content":"

1965年生。博士,教授,博士生导师。主要研究方向为深度学习、点云分析

"}, bioImg=1/mEn+lIJLGJwNIYPV7cgQ==, bioContent=

1965年生。博士,教授,博士生导师。主要研究方向为深度学习、点云分析

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1329420238258548744, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, xref=4, ext=[AuthorCompanyExt(id=1329420238266937353, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238258548744, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=4School of Mathematical Sciences, Zhejiang Normal University, Jinhua, Zhejiang 321014), AuthorCompanyExt(id=1329420238271131658, tenantId=1146029695717560320, journalId=1324759914146529478, articleId=1329420235100238821, companyId=1329420238258548744, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=4浙江师范大学数学科学学院 浙江金华 321014)])])]
邱佳悦1 , 叶海良1, 梁吉业2, 吕科3, 曹飞龙4, *
作者信息
  • 1中国计量大学理学院 杭州 310018
  • 2计算智能与中文信息处理教育部重点实验室(山西大学) 太原 030006
  • 3中国科学院大学工程科学学院 北京 101408
  • 4浙江师范大学数学科学学院 浙江金华 321014
通讯作者:
作者简介:

2001年生。硕士研究生。主要研究方向为深度学习、点云分析

1990年生。博士,副教授,硕士生导师。CCF会员。主要研究方向为深度学习、点云分析

1962年生。博士,教授,博士生导师。CCF会士。主要研究方向为人工智能、粒计算、数据挖掘

1971年生。博士,教授,博士生导师。CCF会员。主要研究方向为计算机视觉、计算机图形学

1965年生。博士,教授,博士生导师。主要研究方向为深度学习、点云分析

A Geometric Point Cloud Completion Network Based on Hypergraph Convolution
Jiayue Qiu1 , Hailiang Ye1, Jiye Liang2, Ke Lü3, Feilong Cao4, *
Affiliations
  • 1College of Sciences, China Jiliang University, Hangzhou 310018
  • 2Key Laboratory of Computational Intelligence and Chinese Information Processing (Shanxi University), Ministry of Education, Taiyuan030006
  • 3School of Engineering Science, University of Chinese Academy of Sciences, Beijing 101408
  • 4School of Mathematical Sciences, Zhejiang Normal University, Jinhua, Zhejiang 321014
  • born in 2001. Master candidate. Her main research interests include deep learning and point cloud analysis

    born in 1990. PhD, associate professor, master supervisor. Member of CCF. His main research interests include deep learning and point cloud analysis

    born in 1962. PhD, professor, PhD supervisor. Fellow of CCF. His main research interests include artificial intelligence, granular computing, and data mining

    born in 1971. PhD, professor, PhD supervisor. Member of CCF. His main research interests include computer vision and computer graphics

    born in 1965. PhD, professor, PhD supervisor. His main research interests include deep learning and point cloud analysis

出版时间: 2026-08-01 doi: 10.7544/issn1000-1239.202550794
文章导航
收藏切换

深度学习方法因其具备快速推理和良好泛化能力,在点云补全任务中取得了显著进展。然而,现有方法较少关注点云的局部几何信息,如法线、向量夹角及高阶关系等。为此,提出一种基于超图卷积的几何点云补全网络,以生成细节丰富的完整点云。该框架主要包括2个核心组件:几何特征提取器与基于超图卷积的粗生成模块。具体而言,前者充分提取点云的法线、中心点与采样点之间的差异向量及其夹角,并将其作为位置编码融入Transformer结构中,使编码器更充分地学习点云的几何特性与旋转不变特征。此外,在基于超图卷积的粗生成模块中,通过引入超图建模点之间的高阶相关性,并利用交叉注意力机制探索逐点特征与高阶邻域特征间的交互作用,指导生成更具几何结构合理性的点云。在多个数据集上的大量实验结果验证了该方法具有较为优越的性能。

图神经网络  /  点云补全  /  几何深度学习  /  超图学习  /  特征提取

Deep learning methods have achieved remarkable progress in point cloud completion tasks due to their fast inference and good generalization capabilities. However, existing methods pay less attention to the local geometric information of point clouds, such as normal vectors, vector angles, and higher-order relationships. To address this issue, we propose a geometric point cloud completion network based on hypergraph convolution to generate detailed and complete point clouds. The framework mainly consists of two core components: a geometric feature extractor and a coarse generation module based on hypergraph convolution. Specifically, the former fully extracts the normal vectors, the difference vectors between the center points and the sampled points, and their angles, and integrates them as position encodings into the Transformer structure, enabling the encoder to more fully learn the geometric characteristics and rotation-invariant features of the point cloud. Additionally, in the coarse generation module based on hypergraph convolution, by introducing hypergraph modeling to capture the higher-order correlations between points and using the cross-attention mechanism to explore the interaction between per-point features and higher-order neighborhood features, it guides the generation of point clouds with more geometric structural rationality. Extensive experimental results on multiple datasets verify that the proposed method has superior performance.

graph neural network  /  point cloud completion  /  geometric deep learning  /  hypergraph learning  /  feature extraction
邱佳悦, 叶海良, 梁吉业, 吕科, 曹飞龙. 基于超图卷积的几何点云补全网络. 计算机研究与发展, 2026 , 63 (8) : 2154 -2168 . DOI: 10.7544/issn1000-1239.202550794
Jiayue Qiu, Hailiang Ye, Jiye Liang, Ke Lü, Feilong Cao. A Geometric Point Cloud Completion Network Based on Hypergraph Convolution[J]. Journal of Computer Research and Development, 2026 , 63 (8) : 2154 -2168 . DOI: 10.7544/issn1000-1239.202550794
近年来,3D传感器技术的快速发展极大地推动了3D计算机视觉领域的研究。点云作为一种常见且易于获取的3D数据表示形式,在自动驾驶[1]、机器人学[2]、表面重建[3]等领域得到了广泛应用。然而,受限于遮挡、传感器分辨率不足等因素,实际采集的原始点云往往存在稀疏性和不完整性。为此,点云补全任务应运而生,旨在从不完整的部分点云中恢复出完整的几何形状,为下游任务提供了可靠的3D结构信息,具有重要的研究价值与应用意义。
在过去十几年中,点云补全得到了广泛研究,涌现出众多方法,主要分为传统方法和基于深度学习的方法两大类。传统方法通常依赖于优化形状模型的参数,以拟合给定的部分输入并生成完整的点云。代表性方法包括基于对齐的技术[4]和基于匹配补丁搜索的策略[5]等。然而,这类方法往往需要进行复杂的优化或大规模的搜索过程,导致计算耗时较长,难以满足实时性需求。自Qi等人[6]提出PointNet和PointNet++[7]以来,直接处理3D点云成为主流范式,推动了深度学习在该领域的广泛应用。得益于其高效的推理能力和良好的泛化性能,深度学习方法近年来在点云补全任务中取得了显著进展。Yuan等人[8]率先提出了基于编码器-解码器架构的端到端点云补全网络(point cloud completion network,PCN),为后续研究奠定了基础。在此框架之上,众多改进模型相继被提出。例如,基于生成对抗网络的方法[9-10]引入对抗损失以提升生成点云的局部细节与真实感;基于折叠策略的方法[11-12]从固定大小的2D网格中采样,并将其与编码器提取的全局特征相结合,逐步映射至3D空间以生成完整点云;基于Transformer的方法[13-14]利用注意力机制,实现对长距离依赖关系的建模,有效地捕捉点云结构信息。尽管上述方法显著提升了补全效果,但未充分考虑点云所蕴含的内在几何特性,限制了其重建精度与几何保真度。
最近,图神经网络方法被广泛应用于点云处理。由于点云本质上属于非欧几里得结构化数据,图方法通常将点云中的每个点视为图中的顶点,并通过$ K $NN算法在邻近点之间构建边以捕捉局部拓扑结构。动态图神经网络DGCNN(dynamic graph CNN for learning on point clouds)[15]作为该领域的开创性工作,引入了动态图卷积机制,为点云分类、分割等任务提供了新的思路。受其启发,许多后续方法[16-17]采用DGCNN作为编码器,以提取更具判别性的特征表示,提升了点云补全的精度。为增强模型对局部结构的感知能力,一些研究[18-19]利用图卷积策略显著提升了局部特征的学习效果。
然而,尽管上述方法已取得一定成效,但仍存在若干不足之处:
1)现有的大多数基于学习的点云补全方法主要依赖于点的坐标信息,忽视了其中蕴含的丰富几何先验知识,导致所学特征难以有效捕捉复杂的形状细节,限制了模型的几何表达能力。此外,点云补全的核心目标是恢复物体本身完整的3D结构,而非特定坐标系下的具体姿态。因此,本文在编码器阶段提取多种特定的几何先验特征,充分学习点云固有几何特性,并获得对旋转和平移不变的特征。
2)先前基于图的补全方法主要局限于建模点与点之间的成对关系,难以有效捕捉点云中复杂的高阶相关性,制约了其对高阶结构信息的表达能力。然而,在点云补全任务中,不同局部邻域之间的信息相互引导对于生成合理的几何结构至关重要。超图学习通过建模多个点之间的高阶关联,实现远距离但语义相似邻域间的信息交互,从而促进结构对称性与整体合理性的协同生成。因此,本文同时关注逐点特征和跨邻域特征,充分利用二者之间的差异性与互补性,以更有效地指导补全结果的生成。
为此,本文提出了一种基于超图卷积的几何点云补全网络,其遵循编码器-解码器的基本架构,并采用由粗到细的2阶段策略生成完整且稠密的点云。网络的核心设计包括几何特征提取器和基于超图卷积的粗生成模块。具体而言,几何特征提取器引入几何Transformer结构,通过提取点云的表面法线、中心点与采样点之间的差异向量,及对应的夹角信息,将其作为位置编码输入至Transformer,增强模型对局部几何结构的辨识能力。此外,由于所提取的几何特征对点云旋转和平移的不变性,该编码器能学习具有一定刚体变换鲁棒性的特征表示。此外,本文还设计了基于超图的粗生成模块,利用超图结构捕捉点之间复杂的高阶关系,并将其与编码器提取到的局部几何特征进行交叉注意力融合,重点关注逐点特征与不同邻域内特征之间的差异,从而更充分地学习和挖掘点云的高阶结构信息与局部几何细节,以实现精细的点云补全。在合成数据集和与真实数据集上的实验结果表明,所提出的基于超图的几何点云补全网络实现了更好的补全效果。综上,本文的主要贡献如下:
1)提出了一种基于超图卷积的几何点云补全网络,生成了具有更丰富几何细节的完整点云,有效地提升了点云补全的性能。
2)设计了几何特征编码器,通过提取点云中各点的法线、点间差异向量及其夹角,充分挖掘局部几何特征与刚体变换下的不变特征,增强模型对几何关系的表达能力。
3)构建了基于超图卷积的粗生成模块,引入超图结构以突破传统图结构仅能建模成对关系的局限,有效捕捉点云中点之间的高阶关联性,并以此指导粗糙点云的生成过程。
基于深度学习的3D点云处理技术[20]的快速发展推动了点云补全领域的研究进展。Yuan等人[8]提出了首个基于学习的点云补全框架,采用编码器-解码器结构,直接对原始不完整点云进行处理,输出完整点云。之后,激发了大量在特征提取与生成机制上的改进方法。Huang等人[9]利用多尺度生成网络分层估计缺失点云,并将多阶段补全损失和对抗损失相加,以生成更真实的缺失区域。Wang等人[10]采用级联细化网络来生成详细的点云形状,并应用生成对抗网络约束生成点云的每个局部块。Wen等人[11]提出了一种具有分层折叠机制的解码器以充分利用编码器得到的信息。Zong等人[12]设计了一个具有自适应采样模块的编码器以及具有跳跃注意力和多分辨率补全目标的分层折叠解码器,提升了点云补全的鲁棒性。
近年来,Transformer因其在自然语言处理领域取得了显著优势,逐渐被引入点云补全任务中。Xiang等人[13]提出一种新的跳跃Transformer,用于学习子点与父点之间的形状上下文及空间关系。Zhang等人[14]提出骨架细节Transformer,通过融合交叉注意力与自注意力层,深入挖掘局部形状与全局形状之间的相关性。此外,Yu等人[17]对局部几何关系进行结构建模并提出几何感知Transformer,更好地利用点云的3D结构。Zhou等人[21]将Transformer机制融入生成过程的基本操作,设计上采样Transformer以更好地捕捉相邻点间的空间与语义关联。为预测每个点准确的位移,Wen等人[22]利用Transformer增强网络对点特征的学习能力。Li等人[23]设计了一个缺失部分敏感Transformer,以感知缺失点的位置信息。Wang等人[24]利用自注意力与交叉注意力建立了点之间的关系,实现具有精确几何细节的点云。最近,Rong等人[25]设计了交叉分辨率Transformer,利用局部注意力有效地进行跨分辨率聚合。Liang等人[26]为了获得高质量的解码特征,提出了几何交互Transformer,借助交叉注意力机制捕捉不完整对象和解码后完整查询矩阵之间的语义关系。Fei等人[27]引入拓扑感知模块来增强Transformer架构,提升了点云拓扑信息的利用。 Fang等人[28]提出交错注意力来增强Transformer,实现信息的双向补充。
由于点云和图可被视为非欧几里得结构化数据,将点云中的点作为图的顶点来建模点间关系成为一种自然且有效的方式。Wang等人[15]率先在点云中引入动态图卷积,提出边卷积(EdgeConv)以更好地捕捉局部几何特征。随后,基于图神经网络的方法在点云补全任务中开始应用。Yu等人[16-17]采用轻量化的DGCNN网络作为补全框架的编码器,提取点云特征。 为了增强对局部细节的感知能力,Xiao等人[18]设计了多分辨率多维图编码器来捕获维度内和跨维度的交互信息。Pan[29]通过图卷积实现点特征的上采样,更好地保留了局部几何细节。Shi等人[30]提出图引导变形网络,利用图卷积网络模拟拉普拉斯优化过程。最近,Huang等人[19]设计了缺失到部分图卷积细化器,将部分点的局部几何信息和语义信息传递到缺失点。Du等人[31]为了在更细粒度的补丁生成中包含局部细节,设计了局部生成图模块,利用图方法捕获局部信息。Shu等人[32]开发了几何感知图集成模块以提高结构一致性。
本文提出了一种基于超图卷积的几何点云补全网络,旨在充分学习点云的局部几何特征及增强特征的旋转平移不变性。如图1所示,整体网络框架采用编码器-解码器框架,主要由几何特征提取器模块和基于超图的粗生成模块组成。首先,设$ {\boldsymbol{P}}_{0}= ({\boldsymbol{p}}_{i})_{i=1}^{{N}_{0}}\in {\mathbb{R}}^{{{N}_{0}}\times 3} $为输入的缺失点云,其中$ {N}_{0} $是点的数量,每一个点$ {\boldsymbol{p}}_{i} $表示一个3D坐标。
在编码阶段,本文构建了一个几何特征提取器(geometric feature extractor,GFE)来编码点云的几何特性。该模块提取点云的局部特征$ {\hat{\boldsymbol{F}}}_{2}=({\hat{\boldsymbol{f}}}_{i})_{i=1}^{N'_{2}}\in {\mathbb{R}}^{{N'_{2}}\times {{C}_{2}}} $、对应的坐标$ {\hat{\boldsymbol{P}}}_{2}=({\hat{\boldsymbol{p}}}_{i})_{i=1}^{N'_{2}}\in {\mathbb{R}}^{{N'_{2}}\times 3} $以及全局特征$ \boldsymbol{g}\in {\mathbb{R}}^{1\times c} $,表示为
$ \{{\hat{\boldsymbol{P}}}_{2},{\hat{\boldsymbol{F}}}_{2},\boldsymbol{g} \}=GFE\left({\boldsymbol{P}}_{0}\right)\;。 $
在解码阶段,采用由粗到细的2阶段法来生成点云。对于粗生成阶段,本文构建了基于超图卷积的粗生成模块(coarse generation module based on hypergraph convolution,CGMHC),得到粗糙但完整的点云$ {\boldsymbol{P}}_{1}\in {\mathbb{R}}^{{{N}_{1}}\times 3} $。该过程可描述为
$ {\boldsymbol{P}}_{1}=CGMHC ({\hat{\boldsymbol{P}}}_{2},{\hat{\boldsymbol{F}}}_{2},\boldsymbol{g} )\;。 $
随后,为了生成更高分辨率的点云,将编码器得到的全局特征$ \boldsymbol{g} $和由$ CGMHC(\cdot ) $生成的粗糙点云$ {\boldsymbol{P}}_{1} $输入到细化模块(refinement module,RM)中,通过3个级联的细化模块得到精细且完整的点云$ {\boldsymbol{P}}_{2}\in {\mathbb{R}}^{{{N}_{2}}\times 3} $。该过程可以被建模为
$ {\boldsymbol{P}}_{2}=RM\left({\boldsymbol{P}}_{1},\boldsymbol{g}\right)\;, $
其中$ RM(\cdot ) $遵循CRA-PCN中的“Up-sampling block”, 其具体实现可以参考文献[25]。由于篇幅限制,此处不再赘述。
本文重点介绍GFE和CGMHC的设计。
点云通常蕴含丰富的几何先验信息,如法线方向、点间差异向量以及差异向量之间的夹角[33]等,这些信息描述了表面局部形状、点之间精确的相对位置关系以及局部几何结构。这些几何特征在点云经历旋转或平移变换时应当保持不变。因此,在编码器中充分提取并利用此类几何信息,不仅有助于更准确地建模点云的几何结构与拓扑关系,还能增强对旋转和平移的不变性,指导生成更具几何合理性且细节完整的点云。 然而,现有的工作通常只为Transformer提供坐标特征,缺乏显式的编码几何结构,使得学习到的特征在几何上的辨别力较弱,难以捕捉其几何不变性。为此,本文提出了几何特征提取。
如图2(b)所示,本文主要探索点云的法线、点间差异向量及差异向量间的角度3类几何先验信息。对于输入点云$ {\boldsymbol{P}}_{0} $,以$ {\boldsymbol{p}}_{i}\in {\boldsymbol{P}}_{0} $为中心点,利用球查询得到$ m $个邻居点$ {\boldsymbol{p}}_{j} $。
1)法线信息。首先,计算邻域内所有点的质心$ \bar{\boldsymbol{p}} $,表示为
$ \bar{\boldsymbol{p}}=\frac{1}{m}\displaystyle\sum_{j=1}^{m}\,{\boldsymbol{p}}_{j}\;, $
接着,计算局部邻域的协方差矩阵
$ {\mathcal{C}} =\frac{1}{m}\displaystyle\sum_{j=1}^{m}\,\left({\boldsymbol{p}}_{j}-\bar{\boldsymbol{p}}\right)\cdot ({\boldsymbol{p}}_{j}-\bar{\boldsymbol{p}})^{\text{T}}\;, $
然后,对协方差矩阵进行特征值分解:
$ \boldsymbol{C}\cdot {\boldsymbol{v}}_{l}={\lambda }_{l}\cdot {\boldsymbol{v}}_{l}\;, $
其中$ {\lambda }_{l} $是特征值,$ {\boldsymbol{v}}_{l} $是对应的特征向量,最小特征值对应的特征向量即法线。
2)差异向量信息。设$ {\boldsymbol{F}}_{\text{dis}}({\boldsymbol{p}}_{i,j}) $表示为中心点$ {\boldsymbol{p}}_{i} $和邻居点$ {\boldsymbol{p}}_{j} $之间的差异向量,即
$ {\boldsymbol{F}}_{\text{dis}} ({\boldsymbol{p}}_{i,j})={\boldsymbol{p}}_{i}-{\boldsymbol{p}}_{j}\;。 $
为了描述中心点$ {\boldsymbol{p}}_{i} $和每个邻居点$ {\boldsymbol{p}}_{j} $之间的曲面,引入特征$ {\boldsymbol{F}}_{\text{ppf}}({\boldsymbol{p}}_{i,j}) $,表示为
$ \left\{\begin{aligned}& {\boldsymbol{F}}_{\text{ppf}}({\boldsymbol{p}}_{i,j})=concat\left({\boldsymbol{F}}_{1},{\boldsymbol{F}}_{2},{\boldsymbol{F}}_{3},{\boldsymbol{F}}_{4}\right)\;,\\&{\boldsymbol{F}}_{1}=\mathrm{\angle }\left({\boldsymbol{n}}_{i},{\boldsymbol{F}}_{\text{dis}}({\boldsymbol{p}}_{i,j})\right)\;,\\&{\boldsymbol{F}}_{2}=\mathrm{\angle }\left({\boldsymbol{n}}_{j},{\boldsymbol{F}}_{\text{dis}}({\boldsymbol{p}}_{i,j})\right)\;,\\&{\boldsymbol{F}}_{3}=\mathrm{\angle }({\boldsymbol{n}}_{i},{\boldsymbol{n}}_{j})\;,\\&{\boldsymbol{F}}_{4}=\|{\boldsymbol{F}}_{\text{dis}}({\boldsymbol{p}}_{i,j})\|\;,\end{aligned}\right. $
其中$ concat(\cdot ) $表示拼接操作,$ {\boldsymbol{n}}_{i} $和$ {\boldsymbol{n}}_{j} $分别表示$ {\boldsymbol{p}}_{i} $,$ {\boldsymbol{p}}_{j} $的表面法线,$ \|\cdot \| $表示欧氏距离,$ \angle ({\cdot }_{1},{\cdot }_{2}) $计算如下:
$ \mathrm{\angle }\left({\cdot }_{1},{\cdot }_{2}\right)=atan2\left(\| {\cdot }_{1}\times {\cdot }_{2}\| ,\left({\cdot }_{1}\right)\cdot \left({\cdot }_{2}\right)\right)\;, $
其中$ atan2(\cdot ) $计算2个向量间的弧度,“$ \times $”和“$ \cdot $”分别表示向量的叉积与点积。由于2个向量之间的欧式距离不会随着其旋转而发生改变,而向量的点积操作同样不会发生改变。因此,得到的弧度值同样也不会发生改变, 从而提取到的特征$ {\boldsymbol{F}}_{\text{ppf}}({\boldsymbol{p}}_{i,j}) $对点云的旋转和平移具有不变性。
3)差异向量之间的角度信息。计算中心点邻域内的差异向量和整个点云内的最大长度的差异向量之间的角度描述点云的几何结构,表示为
$ {\boldsymbol{F}}_{\text{ang}}({\boldsymbol{p}}_{i,j})=\angle \left({\boldsymbol{D}}_{\max },{\boldsymbol{F}}_{\text{dis}}({\boldsymbol{p}}_{i,j})\right)\;, $
其中$ {\boldsymbol{D}}_{\max } $表示整个点云内最大长度的差异向量。
综上,将上述3个先验描述符和中心点坐标进行拼接并通过MLP$ (\cdot ) $得到差异特征$ {\boldsymbol{F}}_{\text{df}} $,即
$ {\boldsymbol{F}}_{\text{df}}=MLP(concat({\boldsymbol{p}}_{i},{\boldsymbol{F}}_{\text{dis}}({\boldsymbol{p}}_{i,j}),{\boldsymbol{F}}_{\text{ppf}}({\boldsymbol{p}}_{i,j}),{\boldsymbol{F}}_{\text{ang}}({\boldsymbol{p}}_{i,j})))\;。 $
之后,将差异特征$ {\boldsymbol{F}}_{\text{df}} $和前一个集抽象层提取得到的特征$ \boldsymbol{F}'_{i} $作为输入,通过线性变换得到对应的位置、查询、键和值矩阵,如图2(a)所示,表达为
$ \left\{\begin{aligned} & \boldsymbol{R}=Linear\left({\boldsymbol{F}}_{\rm df}\right)\;,\\& \boldsymbol{Q}=Linear\left(\boldsymbol{F}'_{i}\right)\;,\\& \boldsymbol{K}=Linear\left(\boldsymbol{F}'_{i}\right)\;,\\& \boldsymbol{V}=Linear\left(\boldsymbol{F}'_{i}\right)\;,\end{aligned}\right. $
其中Linear$ (\cdot ) $表示线性变换,$ \boldsymbol{R} $表示位置矩阵,$ \boldsymbol{Q} $表示查询矩阵,$ \boldsymbol{K} $表示键矩阵,$ \boldsymbol{V} $表示值矩阵。接着,$ \boldsymbol{R} $,$ \boldsymbol{Q} $,$ \boldsymbol{K} $,$ \boldsymbol{V} $被输入给多头注意力层,描述为
$ \boldsymbol{H}=softmax\left(\frac{\boldsymbol{Q}{\boldsymbol{K}}^{\text{T}}-\boldsymbol{Q}{\boldsymbol{R}}^{\text{T}}}{\sqrt{{{d}}_{i}}}\right)\cdot \boldsymbol{V}\;, $
其中$ \boldsymbol{H} $为多头注意力层的输出,$ {{d}}_{i} $表示第$ i $个注意力头的维度。故整个几何Transformer的输出$ {\hat{\boldsymbol{F}}}_{i} $可用式(14)描述。
$ \left\{\begin{aligned} & \tilde{\boldsymbol{H}}=LayerNorm\left(Linear\left(\boldsymbol{H}\right)+\boldsymbol{V}\right)\;,\\& {\hat{\boldsymbol{F}}}_{i}=LayerNorm\left(Linear\left(\tilde{\boldsymbol{H}}\right)+\tilde{\boldsymbol{H}}\right)\;,\end{aligned}\right. $
其中LayerNorm$ (\cdot ) $表示层归一化操作。
如图3所示,整个几何特征提取器模块可被描述为
$ \left\{\begin{aligned} & \left\{\boldsymbol{P}'_{1},\boldsymbol{F}'_{1}\right\}=S A\left({\boldsymbol{P}}_{0}\right)\;,\\& \left\{{\hat{\boldsymbol{P}}}_{1},{\hat{\boldsymbol{F}}}_{1}\right\}=GeoTr\left(\boldsymbol{P}'_{1},\boldsymbol{F}'_{1}\right)\;,\\& \left\{\boldsymbol{P}'_{2},\boldsymbol{F}'_{2}\right\}=S A\left({\hat{\boldsymbol{P}}}_{1},{\hat{\boldsymbol{F}}}_{1}\right)\;,\\& \left\{{\hat{\boldsymbol{P}}}_{2},{\hat{\boldsymbol{F}}}_{2}\right\}=GeoTr\left(\boldsymbol{P}'_{2},\boldsymbol{F}'_{2}\right)\;,\\&{\boldsymbol{g}} =maxpool\left({\hat{\boldsymbol{F}}}_{2}\right)\;,\end{aligned}\right. $
其中SA$ (\cdot ) $表示PointNet++中的集抽象层[7],GeoTr$ (\cdot ) $表示所构建的几何Transformer,maxpool$ (\cdot ) $表示最大池化。
总之,几何特征提取器在每个集抽象层后添加一个几何Transformer,将法线、点之间的差异向量、差异向量之间的角度替代普通Transformer中显式位置嵌入,增强其感知几何特征的能力。此外,由于每对点之间的距离及每对差异向量之间的角度保持不变,故提取的特征具有良好的几何不变性。
由于点云具有复杂的局部几何结构,识别并理解哪些局部区域在结构上具有关联性显得尤为关键。不同邻域之间的信息交互对于指导缺失点云的完整重建具有重要意义。然而,现有基于图的点云补全方法大多关注点之间的成对关系,忽略了邻域间高阶结构信息的利用。这一局限源于传统图卷积难以建模点的高阶依赖关系。相比之下,超边能够自然地连接多个相关点,更好地刻画高阶关联性。为此,本文引入超图卷积机制[34],以捕捉点云中复杂的结构模式。在此基础上,设计了一个基于超图的粗略生成模块,该模块不仅建模点之间的高阶关系,还通过关注逐点特征与不同邻域信息之间的差异,有效引导网络生成更加合理的点云。
1)超图卷积算子。设$ G=\left\{V,E,\boldsymbol{W}\right\} $表示一个超图,其中$ V $是顶点集,$ E $是超边集,每个超边都有对角矩阵$ \boldsymbol{W} $分配权重。超图$ G $可用$ \left| V\right| \times \left| E\right| $的关联矩阵$ \boldsymbol{H} $来表示,定义为
$ h\left(v,e\right)=\left\{\begin{aligned} & 0,&\mathrm{if}\;v\notin e\;,\\&1,&\mathrm{if}\;v\in e\;,\end{aligned}\right. $
顶点$ v\in V $的度表示为$ d(v)=\displaystyle\sum_{e\in E}w(e)h(v,e) $,边$ e\in E $的度表示为$ \delta (e)=\displaystyle\sum_{v\in V}h(v,e) $。给定节点特征矩阵$ \boldsymbol{X}\in {\mathbb{R}}^{N\times {{C}_{1}}} $,则超图卷积算子可表示为
$ \boldsymbol Y =\sigma \left(\boldsymbol{D}_{v}^{-\tfrac{1}{2}}\boldsymbol{H}\boldsymbol{W}\boldsymbol{D}_{e}^{-1}{\boldsymbol{H}}^{\text{T}}\boldsymbol{D}_{v}^{-\tfrac{1}{2}}\boldsymbol{X}\boldsymbol \varTheta\right)\;, $
其中$ \boldsymbol{Y}\in {\mathbb{R}}^{N\times {{C}_{2}}} $表示更新后的节点特征矩阵,$ {\boldsymbol{D}}_{e} $和$ {\boldsymbol{D}}_{v} $分别表示边度和顶点度的对角矩阵,$ \boldsymbol \varTheta\in {\mathbb{R}}^{{{C}_{1}}\times {{C}_{2}}} $是可训练参数,$ \sigma (\cdot ) $是激活函数。
2)基于超图的粗生成模块。如图4所示,该模块主要包含以下步骤。首先,对编码器得到的全局特征$ \boldsymbol{g} $进行反卷积得到中间特征$ {\boldsymbol{F}'}\in {\mathbb{R}}^{{N'_{2}}\times {{C}_{2}}} $,即
$ {\boldsymbol{F}'}=Deconv(\boldsymbol{g})\;, $
其中$ Deconv(\cdot) $表示反卷积操作。
随后,将$ {\boldsymbol{F}'} $作为初始节点特征,利用$ K $NN算法对其构造超边。采用式(17)所提到的超图卷积算子来学习点之间的高阶关系,表示为
$ {\boldsymbol{F}''}=\sigma \left(\boldsymbol{D}_{v}^{-\tfrac{1}{2}}\boldsymbol{H}\boldsymbol{W}\boldsymbol{D}_{e}^{-1}{\boldsymbol{H}}^{\text{T}}\boldsymbol{D}_{v}^{-\tfrac{1}{2}}{\boldsymbol{F}'}\boldsymbol \varTheta\right)\;, $
其中$ {\boldsymbol{F}''} $表示更新后的节点特征。
为了更好地关注逐点信息和不同邻域内的信息之间的差异,对其作差并利用交叉注意力机制进行信息交互。分别利用编码器得到的特征$ {\hat{\boldsymbol{F}}}_{2} $和上一步得到的特征$ {\boldsymbol{F}''} $来获取查询矩阵$ \boldsymbol{Q}=({\boldsymbol{a}}_{i})_{i=1}^{N'_{2}} $和键矩阵$ \boldsymbol{K}=({\boldsymbol{b}}_{i})_{i=1}^{N'_{2}} $。 此外,将二者组合以获取值矩阵$ \boldsymbol{V}=({\boldsymbol{c}}_{i})_{i=1}^{N'_{2}} $,表述如下
$ \left\{\begin{aligned} & \boldsymbol{Q}=MLP\left({\hat{\boldsymbol{F}}}_{2}\right)\;,\\&\boldsymbol{K}=MLP({\boldsymbol{F}''})\;,\\&\boldsymbol{V}=MLP\left(concat\left({\hat{\boldsymbol{F}}}_{2},{\boldsymbol{F}''}\right)\right)\;。\end{aligned}\right. $
之后,通过注意力矩阵计算逐点特征与不同邻域之间的关系,确保点云的全局形状并生成更多的局部细节,具体计算如下
$ {\boldsymbol{a}}_{ij}=\frac{\exp ({\boldsymbol{a}}_{i}-{\boldsymbol{b}}_{j}+{\boldsymbol{p}\boldsymbol{o}\boldsymbol{s}}_{i})}{\displaystyle\sum_{j\in \mathcal{N}\left(i\right)}\mathrm{exp}({\boldsymbol{a}}_{i}-{\boldsymbol{b}}_{j}+{\boldsymbol{p}\boldsymbol{o}\boldsymbol{s}}_{i})}\;, $
其中$ \mathcal{N}(i) $表示中心点的邻域,$ {\boldsymbol{p}\boldsymbol{o}\boldsymbol{s}}_{i} $表示逐点特征所对应点云坐标$ {\hat{\boldsymbol{P}}}_{2} $的中心点和邻域点之间的差,即$ {\boldsymbol{p}\boldsymbol{o}\boldsymbol{s}}_{i}= {\hat{\boldsymbol{p}}}_{i}-{\hat{\boldsymbol{p}}}_{j} $。然后,将注意力矩阵和值矩阵相乘得到最终输出$ \boldsymbol{O}=({\boldsymbol{o}}_{i})_{i=1}^{N'_{2}} $,表示为
$ {\boldsymbol{o}}_{i}=\displaystyle\sum_{j\in \mathcal{N}\left(i\right)}{\boldsymbol{a}}_{ij}* ({\boldsymbol{c}}_{i}+{\boldsymbol{p}\boldsymbol{o}\boldsymbol{s}}_{i} )\;。 $
因此,增强后的特征$ \tilde{\boldsymbol{F}}\in {\mathbb{R}}^{{N'_{2}}\times {{C}_{2}}} $表示为
$ \tilde{\boldsymbol{F}}=MLP\left(\boldsymbol{O}\right)\;。 $
最后,令$ {\hat{\boldsymbol{H}}}_{0}=\tilde{\boldsymbol{F}} $,将更新后的特征$ {\hat{\boldsymbol{H}}}_{i-1} $与全局特征$ \boldsymbol{g} $拼接起来,将其输入至带有跳跃连接的$ MLP(\cdot ) $中,以此生成粗糙点云$ {\boldsymbol{P}}_{1} $,整个过程可表示为
$ \left\{\begin{aligned} & {\boldsymbol{R}}_{i}=concat\left({\hat{\boldsymbol{H}}}_{i-1};repeat\left({\boldsymbol{g}}\right)\right)\;,\\&{\hat{\boldsymbol{H}}}_{i}={f}_{i}\left({\boldsymbol{R}}_{i}\right)+{\boldsymbol{R}}_{i},\;\;i=1{,}2,3\;,\\&\tilde{\boldsymbol{P}}={f}_{4} ({\hat{\boldsymbol{H}}}_{3} )\;,\\&{\boldsymbol{P}}_{1}=FPS\left(concat (\tilde{\boldsymbol{P}},{\boldsymbol{P}}_{0} )\right)\;,\end{aligned}\right. $
其中$ repeat(\cdot ) $表示重复$ \boldsymbol{g} $使其与$ {\hat{\boldsymbol{H}}}_{i-1} $的列大小保持一致,$ {f}_{i}(\cdot )(i=1{,}2,3{,}4) $表示$ MLP(\cdot ) $操作,$ FPS(\cdot ) $表示最远点采样操作。
总之,本节通过引入超图表示,充分挖掘点云中点之间的高阶相关性。同时,进一步探索逐点信息与不同邻域间信息的差异,并利用注意力机制强化这些差异特征,有效提升点的可辨别性,从而有助于生成完整且富含局部几何细节的点云。
本文与CRA-PCN[25]一致,均采用倒角距离(chamfer distance,CD)作为损失。它主要衡量2个点云的相似性,定义为
$ {L}_{CD}\left(\boldsymbol{X},\boldsymbol{Y}\right)=\frac{1}{\left| \boldsymbol{X}\right| }\displaystyle\sum_{\boldsymbol{x}\in \boldsymbol{X}}\,\underset{\boldsymbol{y}\in \boldsymbol{Y}}{\min }\,||\boldsymbol{x}-{\boldsymbol{y}}||_{2}^{2}\\+\frac{1}{\left| \boldsymbol{Y}\right| }\displaystyle\sum_{\boldsymbol{y}\in \boldsymbol{Y}}\,\underset{\boldsymbol{x}\in \boldsymbol{X}}{\min }\,||\boldsymbol{x}-\boldsymbol{y}||_{2}^{2}\;, $
其中$ \boldsymbol{X} $和$ \boldsymbol{Y} $分别表示生成的点云和真实点云。由于本文所提出的网络采用由粗到细的补全框架,因此,最终损失被描述为
$ L={L}_{CD}\left({\boldsymbol{P}}_{1},{\tilde{\boldsymbol{P}}}_{1}\right)+{L}_{CD}\left({\boldsymbol{P}}_{2},{\tilde{\boldsymbol{P}}}_{2}\right)\;, $
其中$ {\boldsymbol{P}}_{1} $和$ {\boldsymbol{P}}_{2} $分别表示生成的粗糙点云和精细化点云,$ {\tilde{\boldsymbol{P}}}_{1} $表示对真实点云$ {\tilde{\boldsymbol{P}}}_{2} $进行下采样所获得的点云,其分辨率与$ {\boldsymbol{P}}_{1} $保持一致。
为了验证基于超图的几何点云补全网络的有效性,本文在合成数据集和真实数据集上都进行了实验,包括PCN[8],ShapeNets扩展数据集[35]以及KITTI[36]。
1)PCN。PCN数据集是从ShapeNets数据集衍生出的8个类别的子集,是评估点云补全方法最常用的基准数据集之一。不完整点云主要从8个角度将完整的点云投影到2.5D深度图像上。因此,每个样本都包括一个完整的点云以及8个不同的部分点云。对于每个完整物体,从网格表面均匀采样16384 个点,应用最远点采样获得包含2048个点的完整点云。PCN数据集主要包含8个类别:飞机、橱柜、汽车、椅子、灯、沙发、桌子和船只。该数据集共包含 28974 个训练样本、800 个验证样本和 1200 个测试样本。
2)ShapeNets扩展数据集。该数据集同样是由ShapeNets数据集衍生出的子集。它主要引入了8个新类别:公交车、床、书架、长凳、吉他、摩托车、滑板和手枪,旨在评估模型在未见类别上的泛化能力。
3)KITTI。该数据集为真实世界数据集,主要通过激光雷达扫描获取。与合成数据集相比,其缺乏真实的完整点云信息。由于KITTI中部分点云是稀疏的,且不存在对应的完整点云,故所有方法都是在PCN数据集的“汽车”类别上训练。KITTI提供了2401个稀疏点云用于测试。
1)评价指标。为了评估补全结果的质量,本文采用广泛使用的CD作为评价标准,通过引入Fei等人[37]所提出的CD的2种变体,即CD-T和CD-P度量2个点云之间的相似度,具体表示为
$ \left\{\begin{aligned} CD\text{-}T\left(\boldsymbol{X},\boldsymbol{Y}\right)=&\frac{1}{\left| \boldsymbol{X}\right| }\displaystyle\sum_{\boldsymbol{x}\in \boldsymbol{X}}\,\underset{\boldsymbol{y}\in \boldsymbol{Y}}{\min }\,||\boldsymbol{x}-\boldsymbol{y}||_{2}^{2}+\\&\frac{1}{\left| \boldsymbol{Y}\right| }\displaystyle\sum_{\boldsymbol{y}\in \boldsymbol{Y}}\,\underset{\boldsymbol{x}\in \boldsymbol{X}}{\min }\,||\boldsymbol{x}-\boldsymbol{y}||_{2}^{2}\;,\\CD\text{-}P\left(\boldsymbol{X},\boldsymbol{Y}\right)=&\frac{1}{2\left| \boldsymbol{X}\right| }\displaystyle\sum_{\boldsymbol{x}\in \boldsymbol{X}}\,\underset{\boldsymbol{y}\in \boldsymbol{Y}}{\min }\,||\boldsymbol{x}-\boldsymbol{y}|{|}_{2}+\\&\frac{1}{2|\boldsymbol{Y}|}\displaystyle\sum_{\boldsymbol{y}\in \boldsymbol{Y}}\,\underset{\boldsymbol{x}\in \boldsymbol{X}}{\min }\,||\boldsymbol{x}-\boldsymbol{y}|{|}_{2}\;,\end{aligned}\right. $
其中$ \boldsymbol{X} $和$ \boldsymbol{Y} $分别表示预测点云和真实点云。此外,$ F\text{-}S core $也常被用于度量2个点云之间的相似度,表述为
$ \left\{\begin{aligned} & F\text{-}S core=\frac{2PR}{P+R}\;,\\& P\left(\epsilon \right)=\frac{1}{\left| \boldsymbol{X}\right| }\displaystyle\sum_{\boldsymbol{x}\in \boldsymbol{X}}\,\left(\underset{\boldsymbol{y}\in \boldsymbol{Y}}{\min }\,||\boldsymbol{x}-\boldsymbol{y}|{|}_{2}\leq \epsilon \right)\;,\\& R\left(\epsilon \right)=\frac{1}{\left| \boldsymbol{Y}\right| }\displaystyle\sum_{\boldsymbol{y}\in \boldsymbol{Y}}\,\left(\underset{\boldsymbol{x}\in \boldsymbol{X}}{\min }\,||\boldsymbol{x}-\boldsymbol{y}|{|}_{2}\leq \epsilon \right)\;,\end{aligned}\right. $
其中$ \epsilon \gt 0 $是一个超参数,在实验中通常被设置为0.0001。此外,由于KITTI缺乏真实标签,本文采用PCN[8]所提出的一致性(Consistency)来评估补全结果。 它通过计算连续帧中同一实例的补全输出之间的平均 CD-T,定义为
$ Consistency=\frac{1}{S-1}\displaystyle\sum_{i=2}^{S}\,CD{\text{-}}T\left({\boldsymbol{X}}^{i-1},{\boldsymbol{X}}^{i}\right)\;, $
其中$ {\boldsymbol{X}}^{i} $是对应于对同一物体进行第$ i $帧采样的第$ i $个预测输出点云,$ S $是总帧数。
2)实现细节。所有实验均在 PyTorch,NVIDIA RTX2080Ti GPU 配置上进行。本文模型使用 Adam 优化器及其默认超参数设置进行优化。初始学习率为 $ \rm 1E-4 $,每 40 个周期衰减 0.7 倍。批量大小为8,总训练周期数为 100。表1列出了所提出的基于超图卷积的几何点云补全网络的其他实施细节。
本文应用CD-T,CD-P,F-Score指标评估PCN数据集上所有方法的性能。本文方法将与以下点云补全方法进行比较,包括PCN[8],FoldingNet[38],TopNet[39],MSN[40],ECG[29],SeedFormer[21],SnowflakeNet[13],SDT[14],PointAttN[24],CRA-PCN[25],ODGNet[41],EINet[42]。定量结果如表2所示,可以发现提出的方法均取得了8个类别上的最优结果,3个指标的平均值分别为4.65,11.823,0.517。 特别地,在CD-P评价指标上,本文方法在所有类别上均达到了最优。另外,和大部分基于Transformer的方法相比,本文方法通过挖掘点云的局部几何信息和特征的旋转不变性,使得模型精度得到了显著提升。
为了直观地验证提出方法的有效性,本文提供了可视化比较,如图5所示。在灯具类别中,相较于对比方法在灯周围出现杂乱的点,本文方法能够生成整体轮廓更清晰的点云。在沙发类别中,本文方法在保持靠背完整的同时,清晰地补出把手的形状。另外,在桌子类别中,只有本文方法能够生成均匀的桌面。
为了进一步验证提出方法的有效性,本文在PCN数据集上训练了所有的方法,并对ShapeNet扩展数据集上新的8个类别进行测试,同样使用CD-T,CD-P,F-Score评估所有方法的性能。实验结果见表3,本文方法在CD-P和F-Score上均取得了最佳结果。
特别地,在F-Score指标上的5个类别取得最优,3个类别为次优,这也充分说明了本文方法的合理性。
另外,我们分析了在ShapeNet扩展数据集上指标未能取得最优性能的潜在原因。如图6所示,由于该数据集中床、书架类别的某些样本与PCN数据集中沙发、橱柜类别较为相似,在测试时容易受到在PCN数据集训练时获得的先验知识的干扰。
为了验证提出方法的泛化能力,本文在KITTI数据集上评估了真实场景下的补全性能。所有方法都在PCN数据集下的汽车类别上进行训练,并在KITTI数据集上测试。使用一致性作为评价指标,实验结果如表4所示。一致性衡量同一辆车在不同帧下输出结果的相似度。实验结果显示本文方法在该度量上取得了最佳结果,充分证明了本文方法在真实数据集上泛化性能良好,可以有效地识别不同帧下的同一辆车。
本节主要验证了所提出的几何特征提取器和基于超图的粗生成模块中关键组件的有效性。所有消融实验均在PCN数据集上进行,实验结果如表5所示。
具体设置如下:针对模型1,将几何特征提取器的几何Transformer用普通Transformer替代,去除超图卷积和交叉注意力;针对模型2,仅利用普通Transformer替换几何Transformer;针对模型3,仅去除超图卷积;针对模型4,仅去除交叉注意力;针对模型5(本文方法),通过对结果的深入分析,可以发现在去掉超图卷积的情况下,CD-T和CD-P分别下降了0.63,0.522,说明高阶邻域特征对于完整精细的点云生成具有较大作用。同样地,用普通Transformer替换几何Transformer同样会导致精度较大幅度的下降,验证了我们引入几何先验信息的合理性。另外,我们给出各核心阶段的推理时间与训练时间。虽然几何特征提取器模块和基于超图卷积的粗生成模块使推理时间分别增加了18.31 s和14.45 s,但这2项核心设计使CD-T下降了0.5和0.63,精度得到了显著提升。
为了能够直观地展现不同模块在生成效果上的作用,我们还对其进行了可视化,如图7所示。与模型2、模型4相比,模型3所生成的飞机中的点最为混乱,机翼后端与飞机尾端杂点最多。同样地,在橱柜类别中,模型3的右侧框架存在部分缺失,这也说明了超图卷积所提取到的高阶信息对于细节的生成至关重要。此外,与模型3、模型4相比,模型2生成的沙发靠背最为稀疏,这说明了几何Transformer能够生成更均匀的点云。
1)超参数讨论。本节探索了超参数的影响,主要探究了在几何Transformer中利用球查询得到的邻居点数量$ m $的选取以及利用$ K $NN算法构造超边时邻居数量$ K $的选取,具体实验结果见表6。本文经验性地选取了$ m=16,32,64 $,当$ m=64 $时,模型取得最佳结果,$ m $变小时,模型精度均会下降。因此,本文选取球查询得到的m=64。此外,本文还讨论了$ K=5, 10,15 $时的情况,当$ K=10 $或K=15时,模型效果均会变差。因此,本文选取$ K=5 $。另外,我们还进行了不同超参数下的耗时对比,当$ m=64 $时,虽然训练时间较久,但其可以实现最快推理,同时其补全效果也是最优的。$ K=5 $同样能够得到相似结论。因此,我们对于超参数的选取也实现了精度与效率的平衡。
2)复杂度分析。为了评估所提出方法的效率,本文在PCN数据集上对参数数量、运算次数以及推理时间进行了实验对比,结果列于表7中。虽然本文方法表现出较高的计算复杂度,但仍在可接受的范围内,并且在补全性能方面较大幅度地超过了其他方法,这表明该方法有效地平衡了性能与计算成本。
本文提出了一种基于超图卷积的几何点云补全网络。该方法通过设计几何特征提取器模块,充分地利用点云的法线、差异向量及其夹角等几何先验信息,有效提升了编码器所提取特征的几何特性与不变性。此外,通过构建基于超图卷积的粗生成模块,建模点云的高阶关系,通过分析逐点特征与其高阶邻域特征之间的差异,实现对局部几何结构的精细刻画,从而指导更精确的补全过程。大量实验结果表明,本文方法在评价指标上均显著优于现有先进方法,验证了其优越的补全性能。未来工作可进一步探索更多潜在的几何特征(如曲率等),并研究超图结构的不同变体,提升模型在复杂场景下的性能。
邱佳悦负责实验设计并撰写论文;叶海良负责提出方法和修改论文;梁吉业提供论文写作指导;吕科完善了论文思路;曹飞龙提出指导意见并修改论文。

参考文献 引证文献
排序方式:
1
Song Ziying, Liu Lin, Jia Feiyang, et al. Robustness-aware 3D object detection in autonomous driving: A review and outlook[J]. IEEE Transactions on Intelligent Transportation Systems, 2024, 25(11): 15407−15436
2
廖瑞杰, 杨绍发, 孟文霞, 等. SegGraph: 室外场景三维点云闭环检测算法[J]. 计算机研究与发展, 2019, 56(2): 338−348
Liao Ruijie, Yang Shaofa, Meng Wenxia, et al. SegGraph: An algorithm for loop-closure detection in outdoor scenes using 3D point clouds[J]. Journal of Computer Research and Development, 2019, 56(2): 338−348 (in Chinese)
3
Sulzer R, Marlet R, Vallet B, et al. A survey and benchmark of automatic surface reconstruction from point clouds[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025, 47(3): 2000−2019
4
Zhang Junming, Chen Weijia, Wang Yuping, et al. Point set voting for partial point cloud analysis[J]. IEEE Robotics and Automation Letters, 2021, 6(2): 596−603
5
Sarkar K, Varanasi K, Stricker D. Learning quadrangulated patches for 3D shape parameterization and completion[C]//Proc of the 5th Int Conf on 3D Vision. Piscataway, NJ: IEEE, 2017: 383−392
6
Qi C R, Su Hao, Mo Kaichun, et al. PointNet: Deep learning on point sets for 3D classification and segmentation[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2017: 77−85
7
Qi C R, Yi Li, Su Hao, et al. PointNet++: Deep hierarchical feature learning on point sets in a metric space[C]//Advances in Neural Information Processing Systems. Cambridge, MA: MIT, 2017: 5105−5114
8
Yuan Wentao, Khot T, Held D, et al. PCN: Point completion network[C]//Proc of the 6th Int Conf on 3D Vision. Piscataway, NJ: IEEE, 2018: 728−737
9
Huang Zitian, Yu Yikuan, Xu Jiawen, et al. PF-Net: Point fractal network for 3D Point cloud completion[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 7659−7667
10
Wang Xiaogang, Ang M H, Lee G H. Cascaded refinement network for point cloud completion[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 787−796
11
Wen Xin, Li Tianyang, Han Zhizhong, et al. Point cloud completion by skip-attention network with hierarchical folding[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2020: 1936−1945
12
Zong Daoming, Sun Shiliang, Zhao Jing. ASHF-Net: Adaptive sampling and hierarchical folding network for robust point cloud completion[C]//Proc of the 35th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2021: 3625−3632
13
Xiang Peng, Wen Xin, Liu Yushen, et al. Snowflake point deconvolution for point cloud completion and generation with skip-Transformer[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(5): 6320−6338
14
Zhang Wenxiao, Zhou Huajian, Dong Zhen, et al. Point cloud completion via skeleton-detail Transformer[J]. IEEE Transactions on Visualization and Computer Graphics, 2023, 29(10): 4229−4242
15
Wang Yue, Sun Yongbin, Liu Ziwei, et al. Dynamic graph CNN for learning on point clouds[J]. ACM Transactions on Graphics, 2019, 38(5): 1−12
16
Yu Xumin, Rao Yongming, Wang Ziyi, et al. PoinTr: Diverse point cloud completion with geometry-aware Transformers[C]//Proc of the 18th IEEE Int Conf on Computer Vision. Piscataway, NJ: IEEE, 2021: 12478−12487
17
Yu Xumin, Rao Yongming, Wang Ziyi, et al. AdaPoinTr: Diverse point cloud completion with adaptive geometry-aware Transformers[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(12): 14114−14130
18
Xiao Haihong, Xu Hongbin, Li Yuqiong, et al. Multi-dimensional graph interactional network for progressive point cloud completion[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 1−12
19
Huang Minming, Fu Yanggeng, Liu Genggeng, et al. GMP-Net: Graph based missing part patching network for point cloud completion[J]. Engineering Applications of Artificial Intelligence, 2024, 130: 107656
20
李娇娇, 孙红岩, 董雨, 等. 基于深度学习的3维点云处理综述[J]. 计算机研究与发展, 2022, 59(5): 1160−1179
Li Jiaojiao, Sun Hongyan, Dong Yu, et al. Survey of 3-dimensional point cloud processing based on deep learning[J]. Journal of Computer Research and Development, 2022, 59(5): 1160−1179 (in Chinese)
21
Zhou Haoran, Cao Yun, Chu Wenqing, et al. SeedFormer: Patch seeds based point cloud completion with upsample Transformer[C]//Proc of the 17th European Conf on Computer Vision. Berlin: Springer, 2022: 416−432
22
Wen Xin, Xiang Peng, Han Zhizhong, et al. PMP-Net++: Point cloud completion by Transformer-enhanced multi-step point moving paths[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(1): 852−867
23
Li Shanshan, Gao Pan, Tan Xiaoyang, et al. ProxyFormer: Proxy alignment assisted point cloud completion with missing part sensitive Transformer[C]//Proc of the IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2023: 9466−9475
24
Wang Jun, Cui Ying, Guo Dongyan, et al. PointAttN: You only need attention for point cloud completion[C]//Proc of the 38th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2024: 5472−5480
25
Rong Yi, Zhou Haoran, Yuan Lixin, et al. CRA-PCN: Point cloud completion with intra- and inter-level cross-resolution Transformers[C]//Proc of the 38th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2024: 4676−4685
26
Liang Hao, He Zhaoshui, Wang Xu, et al. HGG-Net: Hierarchical geometry generation network for point cloud completion[J]. IEEE Transactions on Intelligent Transportation Systems, 2025, 26(9): 12999−13010
27
Fei Ben, Liu Liwen, Luo Tianyue, et al. Point patches contrastive learning for enhanced point cloud completion[J]. IEEE Transactions on Multimedia, 2025, 27: 581−596
28
Fang Chenghao, Liang Jianqing, Liang Jiye, et al. Multi-modal point cloud completion with interleaved attention enhanced Transformer[C]//Proc of the 34th Int Joint Conf on Artificial Intelligence. San Francisco, CA: Morgan Kaufmann, 2025: 963−971
29
Pan Liang. ECG: Edge-aware point cloud completion with graph convolution[J]. IEEE Robotics and Automation Letters, 2020, 5(3): 4392−4398
30
Shi Jieqi, Xu Lingyun, Heng Liang, et al. Graph-guided deformation for point cloud completion[J]. IEEE Robotics and Automation Letters, 2021, 6(4): 7081−7088
31
Du Zhenjiang, Liu Zhitao, Wang Guan, et al. CMNet: Cross-modal coarse-to-fine network for point cloud completion based on patches[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2025, 35(9): 9132−9147
32
Shu Zhenyu, Yao Jian, Xin Shiqing. DFG-PCN: Point cloud completion with degree-flexible point graph[J]. IEEE Transactions on Visualization and Computer Graphics, 2025, 31(12): 1−14
33
Yuan Yongzhe, Wu Yue, Fan Xiaolong, et al. EGST: Enhanced geometric structure Transformer for point cloud registration[J]. IEEE Transactions on Visualization and Computer Graphics, 2024, 30(9): 6222−6234
34
Feng Yifan, You Haoxuan, Zhang Zizhao, et al. Hypergraph neural networks[C]//Proc of the 33rd AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2019: 3558−3565
35
Wu Zhirong, Song Shuran, Khosla A, et al. 3D ShapeNets: A deep representation for volumetric shapes[C]//Proc of IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2015: 1912−1920
36
Wu Weichao, Xu Yongyang, Zeng Ziyin, et al. Dual-attention fusion point cloud completion with spatially aware clustering[J]. Expert Systems with Applications, 2025, 280: 127557
37
Fei Ben, Yang Weidong, Chen Wenming, et al. Comprehensive review of deep learning-based 3D point cloud completion processing and analysis[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(12): 22862−22883
38
Yang Yaoqing, Feng Chen, Shen Yiru, et al. FoldingNet: Point cloud auto-encoder via deep grid deformation[C]//Proc of IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2018: 206−215
39
Tchapmi L P, Kosaraju V, Rezatofighi H, et al. TopNet: Structural point cloud decoder[C]//Proc of IEEE Conf on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2019: 383−392
40
Liu Minghua, Sheng Lu, Yang Sheng, et al. Morphing and sampling network for dense point cloud completion[C]//Proc of the 34th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2020: 11596−11603
41
Cai Pingping, Scott D, Li Xiaoguang, et al. Orthogonal dictionary guided shape completion network for point cloud[C]//Proc of the 38th AAAI Conf on Artificial Intelligence. Palo Alto, CA: AAAI, 2024: 864−872
42
Cai Pingping, Zhang Canyu, Shi Lingjia, et al. EINet: Point cloud completion via extrapolation and interpolation[C]//Proc of the 18th European Conf on Computer Vision. Berlin: Springer, 2025: 377−393
2026年第63卷第8期
87
40
0
引用本文
BibTeX
文章信息
doi: 10.7544/issn1000-1239.202550794
  • 接收时间:2025-11-03
  • 首发时间:2026-11-17
  • 出版时间:2026-08-01
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2025-11-03
  • 修回日期:2026-04-15
  • 录用日期:2026-04-30
基金
作者信息
    1中国计量大学理学院 杭州 310018
    2计算智能与中文信息处理教育部重点实验室(山西大学) 太原 030006
    3中国科学院大学工程科学学院 北京 101408
    4浙江师范大学数学科学学院 浙江金华 321014

通讯作者:

参考文献
分享链接
https://castjournals.cast.org.cn/joweb/jsjyjyfz/CN/10.7544/issn1000-1239.202550794
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏