Article(id=1249044010443150108, tenantId=1146029695717560320, journalId=1249024232475115590, issueId=1249044006114628363, articleNumber=null, orderNo=null, doi=10.11834/jig.250029, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=null, receivedDate=1737561600000, receivedDateStr=2025-01-23, revisedDate=1745510400000, revisedDateStr=2025-04-25, acceptedDate=null, acceptedDateStr=null, onlineDate=1775724898205, onlineDateStr=2026-04-09, pubDate=1765814400000, pubDateStr=2025-12-16, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1775724898205, onlineIssueDateStr=2026-04-09, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1775724898205, creator=13041195026, updateTime=1775724898205, updator=13041195026, issue=Issue{id=1249044006114628363, tenantId=1146029695717560320, journalId=1249024232475115590, year='2025', volume='30', issue='12', pageStart='3707', pageEnd='3968', issueExtLink='null', onlineDate='null', pubDate='null', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=1, specialIssue=null, createTime=1775724897161, creator=13041195026, updateTime=1775726353303, updator=13041195026, preIssue=null, nextIssue=null, ext={EN=IssueExt(id=1249050113662984471, tenantId=1146029695717560320, journalId=1249024232475115590, issueId=1249044006114628363, language=EN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=), CN=IssueExt(id=1249050113667178776, tenantId=1146029695717560320, journalId=1249024232475115590, issueId=1249044006114628363, language=CN, specialIssueTitle=, coverIllustrator=null, specialIssueEditor=, specialIssueAbout=)}, issueFiles=null}, startPage=3782, endPage=3803, ext={EN=ArticleExt(id=1249044010870969119, articleId=1249044010443150108, tenantId=1146029695717560320, journalId=1249024232475115590, language=EN, title=Recent progress in rotation-invariant point cloud networks, columnId=1249044010699002654, journalTitle=Journal of Image and Graphics, columnName=Review, runingTitle=null, highlight=null, articleAbstract=

In recent years, deep learning networks for point clouds have achieved remarkable advancements, with their robust semantic understanding capabilities propelling research across the entire field of three-dimensional (3D) computer vision. These advancements have enabled accurate and efficient processing of 3D data, supporting applications in autonomous driving, robotics, remote sensing and mapping, and augmented reality. However, 3D point clouds often exhibit complex transformation symmetries, with rotation being a particularly challenging yet critical factor. The spatial coordinates of point clouds, which are the fundamental input to point cloud networks, undergo substantial changes, resulting in feature output variations. However, the semantic information embedded within point clouds theoretically remains consistent under various rotational transformations. This spatial variability substantially impacts the stability and reliability of conventional point cloud deep learning networks in semantic perception tasks, such as recognition, classification, and segmentation, reducing their effectiveness in real-world scenarios characterized by arbitrary orientations and poses. Early studies primarily relied on rotational data augmentation to enhance the robustness of point cloud networks against rotational variations. While data augmentation can improve generalization to some extent, it falls short of addressing the fundamental issue posed by the infinite and continuous nature of the rotation group. Acknowledging these limitations, an increasing number of researchers have shifted their focus toward designing rotation-invariant point cloud deep learning networks, which aim to mitigate the impact of rotation on feature extraction at the architectural level. Therefore, researchers seek to achieve consistent semantic perception regardless of point cloud orientation, thereby enhancing the applicability of deep learning models in real-world scenarios where data can be encountered in arbitrary poses. This paper presents a comprehensive survey of the current state of research on rotation-invariant point cloud networks. The research background is first outlined to highlight the importance of rotation invariance in 3D vision tasks and the challenges posed by rotational symmetries in point cloud data. Then, a systematic categorization of the prevailing mainstream methods is investigated. Particularly, the rotation-invariant point cloud networks can be broadly classified into the following three categories: 1) geometric-guided rotation-invariant methods: Using the traditional geometric analysis algorithms, these methods extract rotation-invariant geometric representations such as relative distances, angles, local reference frames, and canonical poses. These representations are then integrated into point cloud networks, facilitating learning of high-level semantic features and maintaining robustness to rotational transformations simultaneously. 2) Feature-guided rotation-invariant methods: These methods employ rotation-equivariant point cloud networks to extract point cloud representations that contain shape and pose information. Leveraging the inherent principles of equivariant networks, they subsequently remove the pose information from the rotation-equivariant representations, obtaining rotation-invariant point cloud features. 3) Training-guided rotation-invariant methods: These methods focus on designing sophisticated and highly generalizable rotational data augmentation training schemes, allowing non-rotation-invariant point cloud networks to gradually acquire robustness of rotations and achieve stable performance simultaneously. An in-depth analysis of the core concepts and algorithmic improvements that support these methods is provided for each category. The current research content on this issue and methodologies within the academic community are outlined, and the advantages and disadvantages of each method are summarized and compared. Subsequently, a comprehensive overview of the prevalent downstream tasks in the research of rotation-invariant point cloud networks is presented. These tasks include point cloud classification, point cloud segmentation, and point cloud retrieval. For each of these tasks, an in-depth discussion of the commonly employed datasets and evaluation metrics, which are essential for assessing network performance, is provided. Additionally, the quantitative performance metrics of mainstream rotation-invariant point cloud networks applied to these tasks are summarized and analyzed, offering a comparative perspective on their efficacy and robustness under rotational variations. Afterward, the downstream application prospects of rotation-invariant point cloud deep learning networks, including point cloud self-supervised representation learning, end-to-end point cloud registration, and point cloud completion, are examined and summarized. Finally, an outlook on future developments and research hotspots is presented. In addition to the ongoing development of new rotation-invariant point cloud networks, three primary issues warrant further research: 1) discrimination of effective geometric attributes. Current approaches are limited by the design of geometric attribute extraction algorithms. An in-depth discussion and determination of the effectiveness of different rotation-invariant geometric attributes within deep learning frameworks could yield novel insights and foster the development of innovative strategies to advance this field. 2) Highly integratable rotation-invariant mechanism. On the one hand, existing non-rotation-invariant point cloud networks continue to demonstrate strong performance on aligned data. The challenge lies in incorporating rotation invariance into these networks in a straightforward manner degrading their original performance. This challenge remains a key research topic because seamless integration requires innovative architectural designs and methodological approaches. On the other hand, rotation-invariant point cloud networks should also exhibit simplicity and reusability, enabling their direct application to downstream tasks with minimal adaptation. 3) High computational efficiency in invariant feature extraction modules. Although many existing methods demonstrate commendable performance, they often incur substantial time and computational costs, making it challenging to efficiently process large-scale point cloud data. Therefore, designing more efficient rotation-invariant point cloud networks that maintain robust feature extraction capabilities while minimizing computational overhead is crucial. Addressing the aforementioned challenges will notably enhance the effectiveness and practicality of rotation-invariant point cloud deep learning networks, facilitating their widespread adoption in complex 3D environments. This survey aims to provide researchers in 3D computer vision with a foundational understanding of current methodologies, highlight key challenges, and suggest potential avenues for future research.

, correspAuthors=Jiaqi Yang, authorNote=null, correspAuthorsNote=null, copyrightStatement=null, 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, authorCompany=null, fund=null, authors=null, authorsList=Zhengbao Wang, Zhenxuan Zeng, Xuan Ouyang, Haozhe Chen, Linjie Li, Jiaqi Yang), CN=ArticleExt(id=1249044020090049531, articleId=1249044010443150108, tenantId=1146029695717560320, journalId=1249024232475115590, language=CN, title=旋转不变点云网络研究进展, columnId=1249044010992603937, journalTitle=中国图象图形学报, columnName=综述, runingTitle=null, highlight=null, articleAbstract=

点云深度学习网络取得显著进展,表现出的强大语义理解能力推动着整个三维视觉领域的进步。然而三维点云通常表现出复杂的变换对称性,其中旋转是一个具有挑战性且必要的主题。不同旋转变换下点云的语义信息一致,但空间坐标不同,这影响了常规点云深度学习网络在语义感知方面的稳定性,难以应用到任意姿态的现实场景。早期的研究主要采用旋转数据增强的方式,但由于旋转本身的无限性和连续性,这种简易方案并不能满足需求。因此,越来越多的学者着手研究具有旋转不变属性的点云深度学习网络,在网络设计层面排除旋转对于特征提取的影响。本文对于旋转不变点云网络相关研究进行充分调研,分析其中存在的挑战,并系统整理相关主流方法,依据旋转不变能力获取方式的不同,将其划分为几何旋转不变方法、特征旋转不变方法和训练旋转不变方法。本文详细描述了当前学术界在该问题上的研究内容和方法,总结和对比各类方法的优缺点,并对常用的一些数据集和评价指标进行整理总结。最后,本文调研和总结了旋转不变点云深度学习网络的下游应用前景,并对未来发展和研究热点进行展望。

, correspAuthors=杨佳琪, authorNote=null, correspAuthorsNote=
杨佳琪
, copyrightStatement=null, copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=h3jNhhlHhiHRgRxImoIWeQ==, magXml=XqrbL744uoqXfMYnw1tbTQ==, pdfUrl=null, pdf=NmDQUtRyftO8pF0aIRs6ZA==, pdfFileSize=3877057, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=D2HeI8qhdFT5bRknRj6KaA==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=FGk55T+OWheM4uSkPc9I0g==, mapNumber=null, authorCompany=null, fund=null, authors=

王正宝,男,硕士研究生,主要研究方向为点云表征。E-mail:

杨佳琪,通信作者,男,长聘副教授,主要研究方向为计算资源、数据标签受限条件下的三维配准重建。E-mail:

曾振轩,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

欧阳轩,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

陈昊哲,男,硕士研究生,主要研究方向为非刚性点云配准。E-mail:

李林杰,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

, authorsList=王正宝, 曾振轩, 欧阳轩, 陈昊哲, 李林杰, 杨佳琪)}, authors=[Author(id=1249044022078148635, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=npu-wzb@mail.nwpu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1249044022233337894, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044022078148635, language=EN, stringName=Zhengbao Wang, firstName=Zhengbao, middleName=null, lastName=Wang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1249044022313029677, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044022078148635, language=CN, stringName=王正宝, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1西北工业大学计算机学院,西安710072, bio={"content":"

王正宝,男,硕士研究生,主要研究方向为点云表征。E-mail:

"}, bioImg=null, bioContent=

王正宝,男,硕士研究生,主要研究方向为点云表征。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1249044020605947922, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, xref=1, ext=[AuthorCompanyExt(id=1249044020614336531, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China), AuthorCompanyExt(id=1249044020622725140, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1西北工业大学计算机学院,西安710072)])]), Author(id=1249044022401110067, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=jqyang@mail.nwpu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1249044022560493629, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044022401110067, language=EN, stringName=Zhenxuan Zeng, firstName=Zhenxuan, middleName=null, lastName=Zeng, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1249044022648574023, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044022401110067, language=CN, stringName=曾振轩, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1西北工业大学计算机学院,西安710072, bio={"content":"

杨佳琪,通信作者,男,长聘副教授,主要研究方向为计算资源、数据标签受限条件下的三维配准重建。E-mail:

"}, bioImg=null, bioContent=

杨佳琪,通信作者,男,长聘副教授,主要研究方向为计算资源、数据标签受限条件下的三维配准重建。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1249044020605947922, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, xref=1, ext=[AuthorCompanyExt(id=1249044020614336531, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China), AuthorCompanyExt(id=1249044020622725140, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1西北工业大学计算机学院,西安710072)])]), Author(id=1249044024594731086, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=zengzhenxuan@mail.nwpu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1249044024733143127, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044024594731086, language=EN, stringName=Xuan Ouyang, firstName=Xuan, middleName=null, lastName=Ouyang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1249044024858972253, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044024594731086, language=CN, stringName=欧阳轩, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1西北工业大学计算机学院,西安710072, bio={"content":"

曾振轩,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

"}, bioImg=null, bioContent=

曾振轩,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1249044020605947922, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, xref=1, ext=[AuthorCompanyExt(id=1249044020614336531, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China), AuthorCompanyExt(id=1249044020622725140, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1西北工业大学计算机学院,西安710072)])]), Author(id=1249044025005772901, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=ouyangxuan@mail.nwpu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1249044026616385646, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044025005772901, language=EN, stringName=Haozhe Chen, firstName=Haozhe, middleName=null, lastName=Chen, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1249044026721243254, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044025005772901, language=CN, stringName=陈昊哲, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1西北工业大学计算机学院,西安710072, bio={"content":"

欧阳轩,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

"}, bioImg=null, bioContent=

欧阳轩,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1249044020605947922, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, xref=1, ext=[AuthorCompanyExt(id=1249044020614336531, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China), AuthorCompanyExt(id=1249044020622725140, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1西北工业大学计算机学院,西安710072)])]), Author(id=1249044026821906556, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, orderNo=4, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=caelumchz@gmail.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1249044026909986945, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044026821906556, language=EN, stringName=Linjie Li, firstName=Linjie, middleName=null, lastName=Li, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1249044027002261638, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044026821906556, language=CN, stringName=李林杰, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=null, address=1西北工业大学计算机学院,西安710072, bio={"content":"

陈昊哲,男,硕士研究生,主要研究方向为非刚性点云配准。E-mail:

"}, bioImg=null, bioContent=

陈昊哲,男,硕士研究生,主要研究方向为非刚性点云配准。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1249044020605947922, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, xref=1, ext=[AuthorCompanyExt(id=1249044020614336531, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China), AuthorCompanyExt(id=1249044020622725140, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1西北工业大学计算机学院,西安710072)])]), Author(id=1249044027111313548, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, orderNo=5, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=lilinjie@mail.nwpu.edu.cn, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1249044027203588242, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044027111313548, language=EN, stringName=Jiaqi Yang, firstName=Jiaqi, middleName=null, lastName=Yang, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1249044027274891414, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, authorId=1249044027111313548, language=CN, stringName=杨佳琪, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=*, address=1西北工业大学计算机学院,西安710072, bio={"content":"

李林杰,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

"}, bioImg=null, bioContent=

李林杰,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1249044020605947922, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, xref=1, ext=[AuthorCompanyExt(id=1249044020614336531, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China), AuthorCompanyExt(id=1249044020622725140, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1西北工业大学计算机学院,西安710072)])])], keywords=[Keyword(id=1249044027388137631, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, orderNo=1, keyword=three-dimensional vision), Keyword(id=1249044027480412324, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, orderNo=2, keyword=deep learning), Keyword(id=1249044027555909802, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, orderNo=3, keyword=point cloud network), Keyword(id=1249044027648184493, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, orderNo=4, keyword=rotation invariance), Keyword(id=1249044027723681971, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, orderNo=5, keyword=rotation equivariance), Keyword(id=1249044027803373754, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, orderNo=1, keyword=三维视觉), Keyword(id=1249044027887259838, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, orderNo=2, keyword=深度学习), Keyword(id=1249044027975340228, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, orderNo=3, keyword=点云网络), Keyword(id=1249044028101169354, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, orderNo=4, keyword=旋转不变性), Keyword(id=1249044028159889613, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, orderNo=5, keyword=旋转等变性)], refs=[Reference(id=1249044032958173535, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2015, volume=114, issue=1, pageStart=1, pageEnd=15, url=null, language=null, rfNumber=null, rfOrder=0, authorNames=Aftab K, Hartley R, Trumpf J, journalName=International Journal of Computer Vision, refType=null, unstructuredReference=Aftab KHartley R and Trumpf J. 2015. Lq-closest-point to affine subspaces using the generalized weiszfeld algorithm. International Journal of Computer Vision114(1): 1-15 [DOI: 10.1007/s11263-014-0791-8], articleTitle=Lq-closest-point to affine subspaces using the generalized weiszfeld algorithm, refAbstract=null), Reference(id=1249044033063031139, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=7156, pageEnd=7165, url=null, language=null, rfNumber=null, rfOrder=1, authorNames=Aoki Y, Goforth H, Srivatsan R A, Lucey S, journalName=null, refType=null, unstructuredReference=Aoki YGoforth HSrivatsan R A and Lucey S. 2019. PointNetLK: robust and efficient point cloud registration using PointNet//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, USA: IEEE:7156-7165 [DOI: 10.1109/CVPR.2019.00733], articleTitle=PointNetLK: robust and efficient point cloud registration using PointNet, refAbstract=null), Reference(id=1249044033151111527, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=1534, pageEnd=1543, url=null, language=null, rfNumber=null, rfOrder=2, authorNames=Armeni I, Sener O, Zamir A R, Jiang H L, Brilakis I, Fischer M, Savarese S, journalName=null, refType=null, unstructuredReference=Armeni ISener OZamir A RJiang H LBrilakis IFischer M and Savarese S. 2016. 3D semantic parsing of large-scale indoor spaces//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE:1534-1543 [DOI: 10.1109/CVPR.2016.170], articleTitle=3D semantic parsing of large-scale indoor spaces, refAbstract=null), Reference(id=1249044033222414698, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=3, authorNames=Assaad S, Downey C, Al-Rfou R, Nayakanti N, Sapp B, journalName=null, refType=null, unstructuredReference=Assaad SDowney CAl-Rfou RNayakanti N and Sapp B. 2023. VN-transformer: rotation-equivariant attention for vector neurons [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2206.04176.pdf, articleTitle=VN-transformer: rotation-equivariant attention for vector neurons, refAbstract=null), Reference(id=1249044033318883694, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2025, volume=622, issue=null, pageStart=null, pageEnd=129313, url=null, language=null, rfNumber=null, rfOrder=4, authorNames=Bello S A, Alfasly S, Mao J W, Lu J, Li L, Xu C, Zou Y R, journalName=Neurocomputing, refType=null, unstructuredReference=Bello S AAlfasly SMao J WLu JLi LXu C and Zou Y R. 2025. Geometric edge convolution for rigid transformation invariant features in 3D point clouds. Neurocomputing622: #129313 [DOI: 10.1016/j.neucom.2024.129313], articleTitle=Geometric edge convolution for rigid transformation invariant features in 3D point clouds, refAbstract=null), Reference(id=1249044033411158386, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2015, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=5, authorNames=Chang A X, Funkhouser T, Guibas L, Hanrahan P, Huang Q X, Li Z M, Savarese S, Savva M, Song S R, Su H, Xiao J X, Yi L, Yu F, journalName=null, refType=null, unstructuredReference=Chang A XFunkhouser TGuibas LHanrahan PHuang Q XLi Z MSavarese SSavva MSong S RSu HXiao J XYi L and Yu F. 2015. ShapeNet: an information-rich 3D model repository [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/1512.03012.pdf, articleTitle=ShapeNet: an information-rich 3D model repository, refAbstract=null), Reference(id=1249044033511821687, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=2275, pageEnd=2279, url=null, language=null, rfNumber=null, rfOrder=6, authorNames=Chen B F, Yeh Y M, Lu Y C, journalName=null, refType=null, unstructuredReference=Chen B FYeh Y M and Lu Y C. 2022. CF-Net: complementary fusion network for rotation invariant point cloud completion//Proceedings of 2022 ICASSP IEEE International Conference on Acoustics, Speech and Signal Processing. Singapore, Singapore: IEEE:2275-2279 [DOI: 10.1109/ICASSP43922.2022.9746388], articleTitle=CF-Net: complementary fusion network for rotation invariant point cloud completion, refAbstract=null), Reference(id=1249044033599902073, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=4989, pageEnd=4997, url=null, language=null, rfNumber=null, rfOrder=7, authorNames=Chen C, Li G B, Xu R J, Chen T S, Wang M, Lin L, journalName=null, refType=null, unstructuredReference=Chen CLi G BXu R JChen T SWang M and Lin L. 2019. ClusterNet: deep hierarchical cluster network with rigorously rotation-invariant representation for point cloud analysis//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, USA: IEEE:4989-4997 [DOI: 10.1109/CVPR.2019.00513], articleTitle=ClusterNet: deep hierarchical cluster network with rigorously rotation-invariant representation for point cloud analysis, refAbstract=null), Reference(id=1249044033671205245, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=14509, pageEnd=14518, url=null, language=null, rfNumber=null, rfOrder=8, authorNames=Chen H W, Liu S C, Chen W K, Li H, Hill R, journalName=null, refType=null, unstructuredReference=Chen H WLiu S CChen W KLi H and Hill R. 2021. Equivariant point network for 3D point cloud analysis//Proceedings of 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville, USA: IEEE:14509-14518 [DOI: 10.1109/CVPR46437.2021.01428], articleTitle=Equivariant point network for 3D point cloud analysis, refAbstract=null), Reference(id=1249044033767674239, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=7462, pageEnd=7471, url=null, language=null, rfNumber=null, rfOrder=9, authorNames=Chen R H, Cong Y, journalName=null, refType=null, unstructuredReference=Chen R H and Cong Y. 2022. The devil is in the pose: ambiguity-free 3D rotation-invariant learning via pose-aware convolution//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, USA: IEEE:7462-7471 [DOI: 10.1109/CVPR52688.2022.00732], articleTitle=The devil is in the pose: ambiguity-free 3D rotation-invariant learning via pose-aware convolution, refAbstract=null), Reference(id=1249044033843171715, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=115, pageEnd=127, url=null, language=null, rfNumber=null, rfOrder=10, authorNames=Chen Y, Shi P C, journalName=null, refType=null, unstructuredReference=Chen Y and Shi P C. 2024. Rotation-invariant completion network//Proceedings of the 6th Chinese Conference on Pattern Recognition and Computer Vision. Xiamen, China: Springer:115-127 [DOI: 10.1007/978-981-99-8432-9_10], articleTitle=Rotation-invariant completion network, refAbstract=null), Reference(id=1249044033952223621, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=5418, pageEnd=5427, url=null, language=null, rfNumber=null, rfOrder=11, authorNames=Chen Y Y, Duan L H, Zhao S S, Ding C X, Tao D C, journalName=null, refType=null, unstructuredReference=Chen Y YDuan L HZhao S SDing C X and Tao D C. 2024. Local-consistent transformation learning for rotation-invariant point cloud analysis//Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE:5418-5427 [DOI: 10.1109/CVPR52733.2024.00518], articleTitle=Local-consistent transformation learning for rotation-invariant point cloud analysis, refAbstract=null), Reference(id=1249044035508310409, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=8957, pageEnd=8965, url=null, language=null, rfNumber=null, rfOrder=12, authorNames=Choy C, Park J, Koltun V, journalName=null, refType=null, unstructuredReference=Choy CPark J and Koltun V. 2019. Fully convolutional geometric features//Proceedings of 2019 IEEE/CVF International Conference on Computer Vision. Seoul, Korea (South): IEEE:8957-8965 [DOI: 10.1109/ICCV.2019.00905], articleTitle=Fully convolutional geometric features, refAbstract=null), Reference(id=1249044035604779405, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2016, volume=null, issue=null, pageStart=2990, pageEnd=2999, url=null, language=null, rfNumber=null, rfOrder=13, authorNames=Cohen T, Welling M, journalName=null, refType=null, unstructuredReference=Cohen T and Welling M. 2016. Group equivariant convolutional networks//Proceedings of the 33rd International Conference on Machine Learning. New York, USA: PMLR:2990-2999, articleTitle=Group equivariant convolutional networks, refAbstract=null), Reference(id=1249044035759968655, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=14, authorNames=Cohen T S, Geiger M, Köhler J, Welling M, journalName=null, refType=null, unstructuredReference=Cohen T SGeiger MKöhler J and Welling M. 2018. Spherical CNNs [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/1801.10130.pdf, articleTitle=Spherical CNNs, refAbstract=null), Reference(id=1249044035852243346, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2025, volume=27, issue=null, pageStart=186, pageEnd=197, url=null, language=null, rfNumber=null, rfOrder=15, authorNames=Dai Y, Ying S H, Gao Y, journalName=IEEE Transactions on Multimedia, refType=null, unstructuredReference=Dai YYing S H and Gao Y. 2025. Exploring local and global consistent correlation on hypergraph for rotation invariant point cloud analysis. IEEE Transactions on Multimedia27: 186-197 [DOI: 10.1109/TMM.2024.3521678], articleTitle=Exploring local and global consistent correlation on hypergraph for rotation invariant point cloud analysis, refAbstract=null), Reference(id=1249044035965489556, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2021a, volume=null, issue=null, pageStart=12180, pageEnd=12189, url=null, language=null, rfNumber=null, rfOrder=16, authorNames=Deng C Y, Litany O, Duan Y Q, Poulenard A, Tagliasacchi A, Guibas L J, journalName=null, refType=null, unstructuredReference=Deng C YLitany ODuan Y QPoulenard ATagliasacchi A and Guibas L J. 2021a. Vector neurons: a general framework for SO(3)- equivariant networks//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:12180-12189 [DOI: 10.1109/ICCV48922.2021.01198], articleTitle=Vector neurons: a general framework for SO(3)- equivariant networks, refAbstract=null), Reference(id=1249044036082930072, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2018a, volume=null, issue=null, pageStart=620, pageEnd=638, url=null, language=null, rfNumber=null, rfOrder=17, authorNames=Deng H W, Birdal T, Ilic S, journalName=null, refType=null, unstructuredReference=Deng H WBirdal T and Ilic S. 2018a. PPF-FoldNet: unsupervised learning of rotation invariant 3D local descriptors//Proceedings of the 15th European Conference on Computer Vision. Munich Germany: Springer:620-638 [DOI: 10.1007/978-3-030-01228-1_37], articleTitle=PPF-FoldNet: unsupervised learning of rotation invariant 3D local descriptors, refAbstract=null), Reference(id=1249044036158427545, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2018b, volume=null, issue=null, pageStart=195, pageEnd=205, url=null, language=null, rfNumber=null, rfOrder=18, authorNames=Deng H W, Birdal T, Ilic S, journalName=null, refType=null, unstructuredReference=Deng H WBirdal T and Ilic S. 2018b. PPFNet: global context aware local features for robust 3D point matching//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE:195-205 [DOI: 10.1109/CVPR.2018.00028], articleTitle=PPFNet: global context aware local features for robust 3D point matching, refAbstract=null), Reference(id=1249044036233925020, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2021b, volume=null, issue=null, pageStart=1, pageEnd=6, url=null, language=null, rfNumber=null, rfOrder=19, authorNames=Deng S, Liu B, Dong Q L, Hu Z Y, journalName=null, refType=null, unstructuredReference=Deng SLiu BDong Q L and Hu Z Y. 2021b. Rotation transformation network: learning view-invariant point cloud for classification and segmentation//Proceedings of 2021 IEEE International Conference on Multimedia and Expo. Shenzhen, China: IEEE:1-6 [DOI: 10.1109/ICME51207.2021.9428265], articleTitle=Rotation transformation network: learning view-invariant point cloud for classification and segmentation, refAbstract=null), Reference(id=1249044036334588316, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2010, volume=null, issue=null, pageStart=998, pageEnd=1005, url=null, language=null, rfNumber=null, rfOrder=20, authorNames=Drost B, Ulrich M, Navab N, Ilic S, journalName=null, refType=null, unstructuredReference=Drost BUlrich MNavab N and Ilic S. 2010. Model globally, match locally: efficient and robust 3D object recognition//Proceedings of 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Francisco, USA: IEEE:998-1005 [DOI: 10.1109/CVPR.2010.5540108], articleTitle=Model globally, match locally: efficient and robust 3D object recognition, refAbstract=null), Reference(id=1249044036431057312, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=1986, volume=103, issue=3/4, pageStart=253, pageEnd=264, url=null, language=null, rfNumber=null, rfOrder=21, authorNames=Escobedo M, journalName=Proceedings of the Royal Society of Edinburgh: Section A Mathematics, refType=null, unstructuredReference=Escobedo M. 1986. On a characterisation of Dirac measures. Proceedings of the Royal Society of Edinburgh: Section A Mathematics103(3/4): 253-264 [DOI: 10.1017/S0308210500018904], articleTitle=On a characterisation of Dirac measures, refAbstract=null), Reference(id=1249044036523332000, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=54, pageEnd=70, url=null, language=null, rfNumber=null, rfOrder=22, authorNames=Esteves C, Allen-Blanchette C, Makadia A, Daniilidis K, journalName=null, refType=null, unstructuredReference=Esteves CAllen-Blanchette CMakadia A and Daniilidis K. 2018. Learning SO(3) equivariant representations with spherical CNNs//Proceedings of the 15th European Conference on Computer Vision. Munich, Germany: Springer:54-70 [DOI: 10.1007/978-3-030-01261-8_4], articleTitle=Learning SO(3) equivariant representations with spherical CNNs, refAbstract=null), Reference(id=1249044036611412387, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024a, volume=null, issue=null, pageStart=985, pageEnd=994, url=null, language=null, rfNumber=null, rfOrder=23, authorNames=Fei J J, Deng Z D, journalName=null, refType=null, unstructuredReference=Fei J J and Deng Z D. 2024a. Incorporating rotation invariance with non-invariant networks for point clouds//Proceedings of 2024 International Conference on 3D Vision. Davos, Switzerland: IEEE:985-994 [DOI: 10.1109/3DV62453.2024.00070], articleTitle=Incorporating rotation invariance with non-invariant networks for point clouds, refAbstract=null), Reference(id=1249044036695298469, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024b, volume=57, issue=7, pageStart=null, pageEnd=168, url=null, language=null, rfNumber=null, rfOrder=24, authorNames=Fei J J, Deng Z D, journalName=Artificial Intelligence Review, refType=null, unstructuredReference=Fei J J and Deng Z D. 2024b. Rotation invariance and equivariance in 3D deep learning: a survey. Artificial Intelligence Review57(7): #168 [DOI: 10.1007/s10462-024-10741-2], articleTitle=Rotation invariance and equivariance in 3D deep learning: a survey, refAbstract=null), Reference(id=1249044036766601638, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=1970, pageEnd=1981, url=null, language=null, rfNumber=null, rfOrder=25, authorNames=Fuchs F B, Worrall D E, Fischer V, Welling M, journalName=null, refType=null, unstructuredReference=Fuchs F BWorrall D EFischer V and Welling M. 2020. SE(3)-transformers: 3D roto-translation equivariant attention networks//Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc.:1970-1981, articleTitle=SE(3)-transformers: 3D roto-translation equivariant attention networks, refAbstract=null), Reference(id=1249044036879847847, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=12, issue=null, pageStart=73340, pageEnd=73353, url=null, language=null, rfNumber=null, rfOrder=26, authorNames=Furuya T, journalName=IEEE Access, refType=null, unstructuredReference=Furuya T. 2024. MaskLRF: self-supervised pretraining via masked autoencoding of local reference frames for rotation-invariant 3D point set analysis. IEEE Access12: 73340-73353 [DOI: 10.1109/ACCESS.2024.3404016], articleTitle=MaskLRF: self-supervised pretraining via masked autoencoding of local reference frames for rotation-invariant 3D point set analysis, refAbstract=null), Reference(id=1249044036967928233, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=29, issue=4, pageStart=24, pageEnd=37, url=null, language=null, rfNumber=null, rfOrder=27, authorNames=Gu R B, Wu Q X, Li Y Q, Kang W X, Ng W W Y, Wang Z Y, journalName=IEEE MultiMedia, refType=null, unstructuredReference=Gu R BWu Q XLi Y QKang W XNg W W Y and Wang Z Y. 2022. Enhanced local and global learning for rotation-invariant point cloud representation. IEEE MultiMedia29(4): 24-37 [DOI: 10.1109/MMUL.2022.3151906], articleTitle=Enhanced local and global learning for rotation-invariant point cloud representation, refAbstract=null), Reference(id=1249044037026648490, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=9, issue=6, pageStart=5014, pageEnd=5021, url=null, language=null, rfNumber=null, rfOrder=28, authorNames=Hao W, Zhang W J, Su H N, journalName=IEEE Robotics and Automation Letters, refType=null, unstructuredReference=Hao WZhang W J and Su H N. 2024. RIA-Net: rotation invariant aware 3D point cloud for large-scale place recognition. IEEE Robotics and Automation Letters9(6): 5014-5021 [DOI: 10.1109/LRA.2024.3384887], articleTitle=RIA-Net: rotation invariant aware 3D point cloud for large-scale place recognition, refAbstract=null), Reference(id=1249044037110534571, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2025, volume=185, issue=null, pageStart=null, pageEnd=107223, url=null, language=null, rfNumber=null, rfOrder=29, authorNames=He C, Zhao Z J, Zhang X B, Yu H, Wang R H, journalName=Neural Networks, refType=null, unstructuredReference=He CZhao Z JZhang X BYu H and Wang R H. 2025. RotInv-PCT: rotation-invariant point cloud transformer via feature separation and aggregation. Neural Networks185: #107223 [DOI: 10.1016/j.neunet.2025.107223], articleTitle=RotInv-PCT: rotation-invariant point cloud transformer via feature separation and aggregation, refAbstract=null), Reference(id=1249044037169254829, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=7659, pageEnd=7667, url=null, language=null, rfNumber=null, rfOrder=30, authorNames=Huang Z T, Yu Y K, Xu J W, Ni F, Le X Y, journalName=null, refType=null, unstructuredReference=Huang Z TYu Y KXu J WNi F and Le X Y. 2020. PF-Net: point fractal network for 3D point cloud completion//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE:7659-7667 [DOI: 10.1109/CVPR42600.2020.00768], articleTitle=PF-Net: point fractal network for 3D point cloud completion, refAbstract=null), Reference(id=1249044037236363695, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=8174, pageEnd=8185, url=null, language=null, rfNumber=null, rfOrder=31, authorNames=Kim S, Park J, Han B, journalName=null, refType=null, unstructuredReference=Kim SPark J and Han B. 2020. Rotation-invariant local-to-global representation learning for 3D point cloud//Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc.:8174-8185, articleTitle=Rotation-invariant local-to-global representation learning for 3D point cloud, refAbstract=null), Reference(id=1249044037303472561, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=5, issue=4, pageStart=105, pageEnd=111, url=null, language=null, rfNumber=null, rfOrder=32, authorNames=Kwon S, Hur J H, Kim H, journalName=JMST Advances, refType=null, unstructuredReference=Kwon SHur J H and Kim H. 2023. Deep learning-based point cloud upsampling: a review of recent trends. JMST Advances5(4): 105-111 [DOI: 10.1007/s42791-023-00058-6], articleTitle=Deep learning-based point cloud upsampling: a review of recent trends, refAbstract=null), Reference(id=1249044037391552947, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=4902, pageEnd=4912, url=null, language=null, rfNumber=null, rfOrder=33, authorNames=Lei J H, Deng C Y, Schmeckpeper K, Guibas L, Daniilidis K, journalName=null, refType=null, unstructuredReference=Lei J HDeng C YSchmeckpeper KGuibas L and Daniilidis K. 2023. EFEM: equivariant neural field expectation maximization for 3D object segmentation without scene supervision//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver, Canada: IEEE:4902-4912 [DOI: 10.1109/CVPR52729.2023.00475], articleTitle=EFEM: equivariant neural field expectation maximization for 3D object segmentation without scene supervision, refAbstract=null), Reference(id=1249044037462856117, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=16198, pageEnd=16207, url=null, language=null, rfNumber=null, rfOrder=34, authorNames=Li F R, Fujiwara K, Okura F, Matsushita Y, journalName=null, refType=null, unstructuredReference=Li F RFujiwara KOkura F and Matsushita Y. 2021. A closer look at rotation-invariant deep point cloud analysis//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:16198-16207 [DOI: 10.1109/ICCV48922.2021.01591], articleTitle=A closer look at rotation-invariant deep point cloud analysis, refAbstract=null), Reference(id=1249044037525770679, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=33, issue=11, pageStart=4356, pageEnd=4378, url=null, language=null, rfNumber=null, rfOrder=35, authorNames=Li G B, Zhang R F, Chen C, Lin L, journalName=Journal of Software, refType=null, unstructuredReference=Li G BZhang R FChen C and Lin L. 2022. Rotation-invariant deep hierarchical cluster network for point cloud analysis. Journal of Software33(11): 4356-4378, articleTitle=Rotation-invariant deep hierarchical cluster network for point cloud analysis, refAbstract=null), Reference(id=1249044037584490937, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022a, volume=33, issue=11, pageStart=4356, pageEnd=4378, url=null, language=null, rfNumber=null, rfOrder=36, authorNames=李冠彬, 张锐斐, 陈超, 林倞, journalName=软件学报, refType=null, unstructuredReference=李冠彬, 张锐斐, 陈超, 林倞. 2022a. 基于旋转不变深度层次聚类网络的点云分析. 软件学报33(11): 4356-4378 [DOI: 10.13328/j.cnki.jos.006315], articleTitle=基于旋转不变深度层次聚类网络的点云分析, refAbstract=null), Reference(id=1249044037643211195, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=7269, pageEnd=7275, url=null, language=null, rfNumber=null, rfOrder=37, authorNames=Li J X, Bi Y C, Lee G H, journalName=null, refType=null, unstructuredReference=Li J XBi Y C and Lee G H. 2019. Discrete rotation equivariance for point cloud recognition//Proceedings of 2019 International Conference on Robotics and Automation. Montreal, Canada: IEEE:7269-7275 [DOI: 10.1109/ICRA.2019.8793983], articleTitle=Discrete rotation equivariance for point cloud recognition, refAbstract=null), Reference(id=1249044037714514365, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=36, issue=9, pageStart=111, pageEnd=117, url=null, language=null, rfNumber=null, rfOrder=38, authorNames=Li S Y, Zhang X D, Sun R, Fan Z G, journalName=Journal of Electronic Measurement and Instrumentation, refType=null, unstructuredReference=Li S YZhang X DSun R and Fan Z G. 2022. Point cloud rotation invariant network based on ellipsoid fitting. Journal of Electronic Measurement and Instrumentation36(9): 111-117, articleTitle=Point cloud rotation invariant network based on ellipsoid fitting, refAbstract=null), Reference(id=1249044037827760575, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022b, volume=36, issue=9, pageStart=111, pageEnd=117, url=null, language=null, rfNumber=null, rfOrder=39, authorNames=黎书玉, 张旭东, 孙锐, 范之国, journalName=电子测量与仪器学报, refType=null, unstructuredReference=黎书玉, 张旭东, 孙锐, 范之国. 2022b. 基于椭球拟合的点云旋转不变网络. 电子测量与仪器学报36(9): 111-117 [DOI: 10.13382/j.jemi.B2205143], articleTitle=基于椭球拟合的点云旋转不变网络, refAbstract=null), Reference(id=1249044037886480832, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=28, issue=12, pageStart=4503, pageEnd=4514, url=null, language=null, rfNumber=null, rfOrder=40, authorNames=Li X Z, Li R H, Chen G Y, Fu C W, Cohen-Or D, Heng P A, journalName=IEEE Transactions on Visualization and Computer Graphics, refType=null, unstructuredReference=Li X ZLi R HChen G YFu C WCohen-Or D and Heng P A. 2022. A rotation-invariant framework for deep point cloud analysis. IEEE Transactions on Visualization and Computer Graphics28(12): 4503-4514 [DOI: 10.1109/TVCG.2021.3092570], articleTitle=A rotation-invariant framework for deep point cloud analysis, refAbstract=null), Reference(id=1249044037949395393, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=25, issue=null, pageStart=3432, pageEnd=3442, url=null, language=null, rfNumber=null, rfOrder=41, authorNames=Li Z Z, Li G, Li T H, Liu S, Gao W, journalName=IEEE Transactions on Multimedia, refType=null, unstructuredReference=Li Z ZLi GLi T HLiu S and Gao W. 2023. Semantic point cloud upsampling. IEEE Transactions on Multimedia25: 3432-3442 [DOI: 10.1109/TMM.2022.3160604], articleTitle=Semantic point cloud upsampling, refAbstract=null), Reference(id=1249044038020698563, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=42, authorNames=Liang D, Guo Y C, Zhang S K, Zhang S H, Hall P, Zhang M, Hu S, journalName=null, refType=null, unstructuredReference=Liang DGuo Y CZhang S KZhang S HHall PZhang M and Hu S. 2018. LineNet: a zoomable CNN for crowdsourced high definition maps modeling in urban environments [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/1807.05696.pdf, articleTitle=LineNet: a zoomable CNN for crowdsourced high definition maps modeling in urban environments, refAbstract=null), Reference(id=1249044038108778949, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=9529, pageEnd=9535, url=null, language=null, rfNumber=null, rfOrder=43, authorNames=Lin J J, Rickert M, Knoll A, journalName=null, refType=null, unstructuredReference=Lin J JRickert M and Knoll A. 2021. Deep hierarchical rotation invariance learning with exact geometry feature representation for point cloud classification//Proceedings of 2021 IEEE International Conference on Robotics and Automation. Xi’an, China: IEEE:9529-9535 [DOI: 10.1109/ICRA48506.2021.9561307], articleTitle=Deep hierarchical rotation invariance learning with exact geometry feature representation for point cloud classification, refAbstract=null), Reference(id=1249044038167499207, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=1817, pageEnd=1825, url=null, language=null, rfNumber=null, rfOrder=44, authorNames=Lou Y J, Ye Z L, You Y, Jiang N J, Lu J B, Wang W M, Ma L Z, Lu C W, journalName=null, refType=null, unstructuredReference=Lou Y JYe Z LYou YJiang N JLu J BWang W MMa L Z and Lu C W. 2023. CRIN: rotation-invariant point cloud analysis and rotation estimation via centrifugal reference frame//Proceedings of the 37th AAAI Conference on Artificial Intelligence. Washington, USA: AAAI:1817-1825 [DOI: 10.1609/aaai.v37i2.25271], articleTitle=CRIN: rotation-invariant point cloud analysis and rotation estimation via centrifugal reference frame, refAbstract=null), Reference(id=1249044038247190984, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=18910, pageEnd=18919, url=null, language=null, rfNumber=null, rfOrder=45, authorNames=Luo S T, Li J H, Guan J Q, Su Y F, Cheng C R, Peng J, Ma J Z, journalName=null, refType=null, unstructuredReference=Luo S TLi J HGuan J QSu Y FCheng C RPeng J and Ma J Z. 2022. Equivariant point cloud analysis via learning orientations for message passing//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, USA: IEEE:18910-18919 [DOI: 10.1109/CVPR52688.2022.01836], articleTitle=Equivariant point cloud analysis via learning orientations for message passing, refAbstract=null), Reference(id=1249044038318494154, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=46, authorNames=Ma X, Qin C, You H X, Ran H X, Fu Y, journalName=null, refType=null, unstructuredReference=Ma XQin CYou H XRan H X and Fu Y. 2022. Rethinking network design and local geometry in point cloud: a simple residual MLP framework [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2202.07123.pdf, articleTitle=Rethinking network design and local geometry in point cloud: a simple residual MLP framework, refAbstract=null), Reference(id=1249044038393991628, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=47, authorNames=Mahler J, Liang J, Niyaz S, Laskey M, Doan R, Liu X Y, Ojea J A, Goldberg K, journalName=null, refType=null, unstructuredReference=Mahler JLiang JNiyaz SLaskey MDoan RLiu X YOjea J A and Goldberg K. 2017. Dex-Net 2.0: deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics [EB/OL]. [2017-08-08]. https://arxiv.org/pdf/1703.09312.pdf, articleTitle=Dex-Net 2.0: deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics, refAbstract=null), Reference(id=1249044038482072013, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=44, issue=12, pageStart=9687, pageEnd=9702, url=null, language=null, rfNumber=null, rfOrder=48, authorNames=Marcon M, Spezialetti R, Salti S, Silva L, Di Stefano L, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Marcon MSpezialetti RSalti SSilva L and Di Stefano L. 2022. Unsupervised learning of local equivariant descriptors for point clouds. IEEE Transactions on Pattern Analysis and Machine Intelligence44(12): 9687-9702 [DOI: 10.1109/TPAMI.2021.3126713], articleTitle=Unsupervised learning of local equivariant descriptors for point clouds, refAbstract=null), Reference(id=1249044040033964495, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=5620, pageEnd=5630, url=null, language=null, rfNumber=null, rfOrder=49, authorNames=Melnyk P, Robinson A, Felsberg M, Wadenbäck M, journalName=null, refType=null, unstructuredReference=Melnyk PRobinson AFelsberg M and Wadenbäck M. 2024. TetraSphere: a neural descriptor for O(3)-invariant point cloud analysis//Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE:5620-5630 [DOI: 10.1109/CVPR52733.2024.00537], articleTitle=TetraSphere: a neural descriptor for O(3)-invariant point cloud analysis, refAbstract=null), Reference(id=1249044040105267664, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=510, pageEnd=522, url=null, language=null, rfNumber=null, rfOrder=50, authorNames=Peng Y, Tang L L, Liao Q, Liu Y, Qi S H, Zhang J J, journalName=null, refType=null, unstructuredReference=Peng YTang L LLiao QLiu YQi S H and Zhang J J. 2024. SE(3)- Diffusion: an equivariant diffusion model for 3D point cloud generation//Proceedings of the 15th International Conference on Genetic and Evolutionary Computing. Kaohsiung, China: Springer:510-522 [DOI: 10.1007/978-981-97-0068-4_48], articleTitle=SE(3)- Diffusion: an equivariant diffusion model for 3D point cloud generation, refAbstract=null), Reference(id=1249044040218513874, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=47, pageEnd=56, url=null, language=null, rfNumber=null, rfOrder=51, authorNames=Poulenard A, Rakotosaona M J, Ponty Y, Ovsjanikov M, journalName=null, refType=null, unstructuredReference=Poulenard ARakotosaona M JPonty Y and Ovsjanikov M. 2019. Effective rotation-invariant point CNN with spherical harmonics kernels//Proceedings of 2019 International Conference on 3D Vision. Quebec City, Canada: IEEE:47-56 [DOI: 10.1109/3DV.2019.00015], articleTitle=Effective rotation-invariant point CNN with spherical harmonics kernels, refAbstract=null), Reference(id=1249044040369508819, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2017a, volume=null, issue=null, pageStart=77, pageEnd=85, url=null, language=null, rfNumber=null, rfOrder=52, authorNames=Qi C R, Su H, Mo K, Guibas L J, journalName=null, refType=null, unstructuredReference=Qi C RSu HMo K and Guibas L J. 2017a. PointNet: deep learning on point sets for 3D classification and segmentation//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE:77-85 [DOI: 10.1109/CVPR.2017.16], articleTitle=PointNet: deep learning on point sets for 3D classification and segmentation, refAbstract=null), Reference(id=1249044040486949333, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2017b, volume=null, issue=null, pageStart=5105, pageEnd=5114, url=null, language=null, rfNumber=null, rfOrder=53, authorNames=Qi C R, Yi L, Su H, Guibas L J, journalName=null, refType=null, unstructuredReference=Qi C RYi LSu H and Guibas L J. 2017b. PointNet++: deep hierarchical feature learning on point sets in a metric space//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, USA: Curran Associates Inc.:5105-5114, articleTitle=PointNet++: deep hierarchical feature learning on point sets in a metric space, refAbstract=null), Reference(id=1249044040583418327, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=45, issue=8, pageStart=9806, pageEnd=9821, url=null, language=null, rfNumber=null, rfOrder=54, authorNames=Qin Z, Yu H, Wang C J, Guo Y L, Peng Y X, Ilic S, Hu D W, Xu K, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Qin ZYu HWang C JGuo Y LPeng Y XIlic SHu D W and Xu K. 2023. GeoTransformer: fast and robust point cloud registration with geometric transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence45(8): 9806-9821 [DOI: 10.1109/TPAMI.2023.3259038], articleTitle=GeoTransformer: fast and robust point cloud registration with geometric transformer, refAbstract=null), Reference(id=1249044040654721497, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=593, pageEnd=609, url=null, language=null, rfNumber=null, rfOrder=55, authorNames=Qiu Z F, Li Y H, Wang Y, Pan Y W, Yao T, Mei T, journalName=null, refType=null, unstructuredReference=Qiu Z FLi Y HWang YPan Y WYao T and Mei T. 2022. SPE-Net: boosting point cloud analysis via rotation robustness enhancement//Proceedings of the 17th European Conference on Computer Vision. Tel Aviv, Israel: Springer:593-609 [DOI: 10.1007/978-3-031-20062-5_34], articleTitle=SPE-Net: boosting point cloud analysis via rotation robustness enhancement, refAbstract=null), Reference(id=1249044040751190491, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2009, volume=null, issue=null, pageStart=3212, pageEnd=3217, url=null, language=null, rfNumber=null, rfOrder=56, authorNames=Rusu R B, Blodow N, Beetz M, journalName=null, refType=null, unstructuredReference=Rusu R BBlodow N and Beetz M. 2009. Fast point feature histograms (FPFH) for 3D registration//Proceedings of 2009 IEEE International Conference on Robotics and Automation. Kobe, Japan: IEEE:3212-3217 [DOI: 10.1109/ROBOT.2009.5152473], articleTitle=Fast point feature histograms (FPFH) for 3D registration, refAbstract=null), Reference(id=1249044040826687965, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=46, issue=5, pageStart=3290, pageEnd=3304, url=null, language=null, rfNumber=null, rfOrder=57, authorNames=Shen W, Wei Z H, Ren Q H, Zhang B B, Huang S K, Fan J Q, Zhang Q S, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=Shen WWei Z HRen Q HZhang B BHuang S KFan J Q and Zhang Q S. 2024. Interpretable rotation-equivariant quaternion neural networks for 3D point cloud processing. IEEE Transactions on Pattern Analysis and Machine Intelligence46(5): 3290-3304 [DOI: 10.1109/TPAMI.2023.3346383], articleTitle=Interpretable rotation-equivariant quaternion neural networks for 3D point cloud processing, refAbstract=null), Reference(id=1249044040885408223, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=531, pageEnd=547, url=null, language=null, rfNumber=null, rfOrder=58, authorNames=Shen W, Zhang B B, Huang S K, Wei Z H, Zhang Q S, journalName=null, refType=null, unstructuredReference=Shen WZhang B BHuang S KWei Z H and Zhang Q S. 2020. 3D-rotation-equivariant quaternion neural networks//Proceedings of the 16th European Conference on Computer Vision. Glasgow, UK: Springer:531-547 [DOI: 10.1007/978-3-030-58565-5_32], articleTitle=3D-rotation-equivariant quaternion neural networks, refAbstract=null), Reference(id=1249044040952517089, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=59, authorNames=Son D, Kim J, Son S, Kim B, journalName=null, refType=null, unstructuredReference=Son DKim JSon S and Kim B. 2024. An intuitive multi-frequency feature representation for SO(3)-equivariant networks [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2405.04537.pdf, articleTitle=An intuitive multi-frequency feature representation for SO(3)-equivariant networks, refAbstract=null), Reference(id=1249044041019625955, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=5381, pageEnd=5392, url=null, language=null, rfNumber=null, rfOrder=60, authorNames=Spezialetti R, Stella F, Marcon M, Silva L, Salti S, Di Stefano L, journalName=null, refType=null, unstructuredReference=Spezialetti RStella FMarcon MSilva LSalti S and Di Stefano L. 2020. Learning to orient surfaces by self-supervised spherical CNNs//Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc.:5381-5392, articleTitle=Learning to orient surfaces by self-supervised spherical CNNs, refAbstract=null), Reference(id=1249044041090929125, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=61, authorNames=Su K M, Wu Q X, Cai P P, Zhu X G, Lu X Q, Wang Z Y, Hu K, journalName=null, refType=null, unstructuredReference=Su K MWu Q XCai P PZhu X GLu X QWang Z Y and Hu K. 2024. RI-MAE: rotation-invariant masked AutoEncoders for self-supervised point cloud representation learning [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2409.00353.pdf, articleTitle=RI-MAE: rotation-invariant masked AutoEncoders for self-supervised point cloud representation learning, refAbstract=null), Reference(id=1249044041158037990, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2025, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=62, authorNames=Thengane V, Zhu X T, Bouzerdoum S, Phung S L, Li Y P, journalName=null, refType=null, unstructuredReference=Thengane VZhu X TBouzerdoum SPhung S L and Li Y P. 2025. Foundational models for 3D point clouds: a survey and outlook [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2501.18594.pdf, articleTitle=Foundational models for 3D point clouds: a survey and outlook, refAbstract=null), Reference(id=1249044041229341159, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=504, pageEnd=513, url=null, language=null, rfNumber=null, rfOrder=63, authorNames=Thomas H, journalName=null, refType=null, unstructuredReference=Thomas H. 2020. Rotation-invariant point convolution with multiple equivariant alignments//Proceedings of 2020 International Conference on 3D Vision. Fukuoka, Japan: IEEE:504-513 [DOI: 10.1109/3DV50981.2020.00060], articleTitle=Rotation-invariant point convolution with multiple equivariant alignments, refAbstract=null), Reference(id=1249044041309032936, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2018, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=64, authorNames=Thomas N, Smidt T, Kearnes S, Yang L S, Li L, Kohlhoff K, Riley P, journalName=null, refType=null, unstructuredReference=Thomas NSmidt TKearnes SYang L SLi LKohlhoff K and Riley P. 2018. Tensor field networks: rotation- and translation-equivariant neural networks for 3D point clouds [EB/OL]. [2018-05-18]. https://arxiv.org/pdf/1802.08219.pdf, articleTitle=Tensor field networks: rotation- and translation-equivariant neural networks for 3D point clouds, refAbstract=null), Reference(id=1249044041392919017, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2010, volume=null, issue=null, pageStart=356, pageEnd=369, url=null, language=null, rfNumber=null, rfOrder=65, authorNames=Tombari F, Salti S, Di Stefano L, journalName=null, refType=null, unstructuredReference=Tombari FSalti S and Di Stefano L. 2010. Unique signatures of histograms for local surface description//Proceedings of the 11th European Conference on Computer Vision. Heraklion, Greece: Springer:356-369 [DOI: 10.1007/978-3-642-15558-1_26], articleTitle=Unique signatures of histograms for local surface description, refAbstract=null), Reference(id=1249044041455833578, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=1588, pageEnd=1597, url=null, language=null, rfNumber=null, rfOrder=66, authorNames=Uy M A, Pham Q H, Hua B S, Nguyen T, Yeung S K, journalName=null, refType=null, unstructuredReference=Uy M APham Q HHua B SNguyen T and Yeung S K. 2019. Revisiting point cloud classification: a new benchmark dataset and classification model on real-world data//Proceedings of 2019 IEEE/CVF International Conference on Computer Vision. Seoul, Korea (South): IEEE:1588-1597 [DOI: 10.1109/ICCV.2019.00167], articleTitle=Revisiting point cloud classification: a new benchmark dataset and classification model on real-world data, refAbstract=null), Reference(id=1249044041539719659, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2017, volume=null, issue=null, pageStart=6000, pageEnd=6010, url=null, language=null, rfNumber=null, rfOrder=67, authorNames=Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez A N, Kaiser Ł and Polosukhin I, journalName=null, refType=null, unstructuredReference=Vaswani AShazeer NParmar NUszkoreit JJones LGomez A NKaiser Ł and Polosukhin I. 2017. Attention is all you need//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, USA: Curran Associates Inc.:6000-6010, articleTitle=Attention is all you need, refAbstract=null), Reference(id=1249044041669743084, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=null, issue=null, pageStart=14351, pageEnd=14360, url=null, language=null, rfNumber=null, rfOrder=68, authorNames=Wang R B, Yang Y B, Tao D C, journalName=null, refType=null, unstructuredReference=Wang R BYang Y B and Tao D C. 2022. ART-point: improving rotation robustness of point cloud classifiers via adversarial rotation//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, USA: IEEE:14351-14360 [DOI: 10.1109/CVPR52688.2022.01397], articleTitle=ART-point: improving rotation robustness of point cloud classifiers via adversarial rotation, refAbstract=null), Reference(id=1249044041837515245, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=38, issue=5, pageStart=null, pageEnd=146, url=null, language=null, rfNumber=null, rfOrder=69, authorNames=Wang Y, Sun Y B, Liu Z W, Sarma S E, Bronstein M M, Solomon J M, journalName=ACM Transactions on Graphics, refType=null, unstructuredReference=Wang YSun Y BLiu Z WSarma S EBronstein M M and Solomon J M. 2019. Dynamic graph CNN for learning on point clouds. ACM Transactions on Graphics38(5): #146 [DOI: 10.1145/3326362], articleTitle=Dynamic graph CNN for learning on point clouds, refAbstract=null), Reference(id=1249044041929789934, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024a, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=70, authorNames=Wang Z X, Han X, Liu H X, Li X Z, journalName=null, refType=null, unstructuredReference=Wang Z XHan XLiu H X and Li X Z. 2024a. RIDE: boosting 3D object detection for LiDAR point clouds via rotation-invariant analysis [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2408.15643.pdf, articleTitle=RIDE: boosting 3D object detection for LiDAR point clouds via rotation-invariant analysis, refAbstract=null), Reference(id=1249044042017870319, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024b, volume=154, issue=null, pageStart=null, pageEnd=110624, url=null, language=null, rfNumber=null, rfOrder=71, authorNames=Wang Z X, Yu Y L, Li X Z, journalName=Pattern Recognition, refType=null, unstructuredReference=Wang Z XYu Y L and Li X Z. 2024b. Rethinking local-to-global representation learning for rotation-invariant point cloud analysis. Pattern Recognition154: #110624 [DOI: 10.1016/j.patcog.2024.110624], articleTitle=Rethinking local-to-global representation learning for rotation-invariant point cloud analysis, refAbstract=null), Reference(id=1249044042089173488, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2025, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=72, authorNames=Weijler L, Hermosilla P, journalName=null, refType=null, unstructuredReference=Weijler L and Hermosilla P. 2025. Efficient continuous group convolutions for local SE(3) equivariance in 3D point clouds [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2502.07505.pdf, articleTitle=Efficient continuous group convolutions for local SE(3) equivariance in 3D point clouds, refAbstract=null), Reference(id=1249044042143699441, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2015, volume=null, issue=null, pageStart=1912, pageEnd=1920, url=null, language=null, rfNumber=null, rfOrder=73, authorNames=Wu Z R, Song S R, Khosla A, Yu F, Zhang L G, Tang X O, Xiao J X, journalName=null, refType=null, unstructuredReference=Wu Z RSong S RKhosla AYu FZhang L GTang X O and Xiao J X. 2015. 3D ShapeNets: a deep representation for volumetric shapes//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE:1912-1920 [DOI: 10.1109/CVPR.2015.7298801], articleTitle=3D ShapeNets: a deep representation for volumetric shapes, refAbstract=null), Reference(id=1249044042206614002, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=10448, pageEnd=10457, url=null, language=null, rfNumber=null, rfOrder=74, authorNames=Xu J Y, Tang X, Zhu Y S, Sun J, Pu S L, journalName=null, refType=null, unstructuredReference=Xu J YTang XZhu Y SSun J and Pu S L. 2021. SGMNet: learning rotation-invariant point cloud representations via sorted Gram matrix//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:10448-10457 [DOI: 10.1109/ICCV48922.2021.01030], articleTitle=SGMNet: learning rotation-invariant point cloud representations via sorted Gram matrix, refAbstract=null), Reference(id=1249044042277917171, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2018, volume=18, issue=10, pageStart=null, pageEnd=3337, url=null, language=null, rfNumber=null, rfOrder=75, authorNames=Yan Y, Mao Y X, Li B, journalName=Sensors, refType=null, unstructuredReference=Yan YMao Y X and Li B. 2018. SECOND: sparsely embedded convolutional detection. Sensors18(10): #3337 [DOI: 10.3390/s18103337], articleTitle=SECOND: sparsely embedded convolutional detection, refAbstract=null), Reference(id=1249044042353414644, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2025, volume=87, issue=null, pageStart=null, pageEnd=102915, url=null, language=null, rfNumber=null, rfOrder=76, authorNames=Yang X Y, Yang R, Hu F, Fang L, Zhong W, Chen Z Y, Wang G, journalName=Displays, refType=null, unstructuredReference=Yang X YYang RHu FFang LZhong WChen Z Y and Wang G. 2025. ART-InvRec: an adversarial framework for rotation-invariant 3D object reconstruction. Displays87: #102915 [DOI: 10.1016/j.displa.2024.102915], articleTitle=ART-InvRec: an adversarial framework for rotation-invariant 3D object reconstruction, refAbstract=null), Reference(id=1249044042420523509, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2025, volume=null, issue=null, pageStart=287, pageEnd=303, url=null, language=null, rfNumber=null, rfOrder=77, authorNames=Yao R Z, Du S Y, Cui W T, Tang C H, Yang C W, journalName=null, refType=null, unstructuredReference=Yao R ZDu S YCui W TTang C H and Yang C W. 2025. PARE-Net: position-aware rotation-equivariant networks for robust point cloud registration//Proceedings of the 18th European Conference on Computer Vision. Milan, Italy: Springer:287-303 [DOI: 10.1007/978-3-031-72904-1_17], articleTitle=PARE-Net: position-aware rotation-equivariant networks for robust point cloud registration, refAbstract=null), Reference(id=1249044042491826678, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=14417, pageEnd=14428, url=null, language=null, rfNumber=null, rfOrder=78, authorNames=Yi X Y, Deng J J, Sun Q R, Hua X S, Lim J H, Zhang H W, journalName=null, refType=null, unstructuredReference=Yi X YDeng J JSun Q RHua X SLim J H and Zhang H W. 2023. Invariant training 2D-3D joint hard samples for few-shot point cloud recognition//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision. Paris, France: IEEE:14417-14428 [DOI: 10.1109/ICCV51070.2023.01330], articleTitle=Invariant training 2D-3D joint hard samples for few-shot point cloud recognition, refAbstract=null), Reference(id=1249044042554741239, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=12717, pageEnd=12724, url=null, language=null, rfNumber=null, rfOrder=79, authorNames=You Y, Lou Y J, Liu Q, Tai Y W, Ma L Z, Lu C W, Wang W M, journalName=null, refType=null, unstructuredReference=You YLou Y JLiu QTai Y WMa L ZLu C W and Wang W M. 2020. Pointwise rotation-invariant network with adaptive sampling and 3D spherical voxel convolution//Proceedings of the 34th AAAI Conference on Artificial Intelligence. New York, USA: AAAI:12717-12724 [DOI: 10.1609/aaai.v34i07.6965], articleTitle=Pointwise rotation-invariant network with adaptive sampling and 3D spherical voxel convolution, refAbstract=null), Reference(id=1249044042621850104, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=44, issue=12, pageStart=9489, pageEnd=9502, url=null, language=null, rfNumber=null, rfOrder=80, authorNames=You Y, Lou Y J, Shi R X, Liu Q, Tai Y W, Ma L Z, Wang W M, Lu C W, journalName=IEEE Transactions on Pattern Analysis and Machine Intelligence, refType=null, unstructuredReference=You YLou Y JShi R XLiu QTai Y WMa L ZWang W M and Lu C W. 2022. PRIN/SPRIN: on extracting point-wise rotation invariant features. IEEE Transactions on Pattern Analysis and Machine Intelligence44(12): 9489-9502 [DOI: 10.1109/TPAMI.2021.3130590], articleTitle=PRIN/SPRIN: on extracting point-wise rotation invariant features, refAbstract=null), Reference(id=1249044042697347577, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023a, volume=null, issue=null, pageStart=5384, pageEnd=5393, url=null, language=null, rfNumber=null, rfOrder=81, authorNames=Yu H, Qin Z, Hou J, Saleh M, Li D S, Busam B, Ilic S, journalName=null, refType=null, unstructuredReference=Yu HQin ZHou JSaleh MLi D SBusam B and Ilic S. 2023a. Rotation-invariant transformer for point cloud matching//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver, Canada: IEEE:5384-5393 [DOI: 10.1109/CVPR52729.2023.00521], articleTitle=Rotation-invariant transformer for point cloud matching, refAbstract=null), Reference(id=1249044042768650746, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=57, issue=7, pageStart=4789, pageEnd=4798, url=null, language=null, rfNumber=null, rfOrder=82, authorNames=Yu J, Lin Y, Wang B, Ye Q, Cai J Q, journalName=IEEE Transactions on Geoscience and Remote Sensing, refType=null, unstructuredReference=Yu JLin YWang BYe Q and Cai J Q. 2019. An advanced outlier detected total least-squares algorithm for 3-D point clouds registration. IEEE Transactions on Geoscience and Remote Sensing57(7): 4789-4798 [DOI: 10.1109/TGRS.2019.2893054], articleTitle=An advanced outlier detected total least-squares algorithm for 3-D point clouds registration, refAbstract=null), Reference(id=1249044042852536827, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023b, volume=null, issue=null, pageStart=3313, pageEnd=3321, url=null, language=null, rfNumber=null, rfOrder=83, authorNames=Yu J H, Zhang C Y, Cai W D, journalName=null, refType=null, unstructuredReference=Yu J HZhang C Y and Cai W D. 2023b. Rethinking rotation invariance with point cloud registration//Proceedings of the 37th AAAI Conference on Artificial Intelligence. Washington, USA: AAAI:3313-3321 [DOI: 10.1609/aaai.v37i3.25438], articleTitle=Rethinking rotation invariance with point cloud registration, refAbstract=null), Reference(id=1249044042928034300, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=217, pageEnd=233, url=null, language=null, rfNumber=null, rfOrder=84, authorNames=Yu R X, Wei X, Tombari F, Sun J, journalName=null, refType=null, unstructuredReference=Yu R XWei XTombari F and Sun J. 2020. Deep positional and relational feature learning for rotation-invariant point cloud analysis//Proceedings of the 16th European Conference on Computer Vision. Glasgow, UK: Springer:217-233 [DOI: 10.1007/978-3-030-58607-2_13], articleTitle=Deep positional and relational feature learning for rotation-invariant point cloud analysis, refAbstract=null), Reference(id=1249044044479926781, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=12478, pageEnd=12487, url=null, language=null, rfNumber=null, rfOrder=85, authorNames=Yu X M, Rao Y M, Wang Z Y, Liu Z Y, Lu J W, Zhou J, journalName=null, refType=null, unstructuredReference=Yu X MRao Y MWang Z YLiu Z YLu J W and Zhou J. 2021. PoinTr: diverse point cloud completion with geometry-aware transformers//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:12478-12487 [DOI: 10.1109/ICCV48922.2021.01227], articleTitle=PoinTr: diverse point cloud completion with geometry-aware transformers, refAbstract=null), Reference(id=1249044044568007166, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=null, rfOrder=86, authorNames=Yuan W T, Held D, Mertz C, Hebert M, journalName=null, refType=null, unstructuredReference=Yuan W THeld DMertz C and Hebert M. 2019. Iterative transformer network for 3D point cloud [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/1811.11209.pdf, articleTitle=Iterative transformer network for 3D point cloud, refAbstract=null), Reference(id=1249044044664476159, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=3418, pageEnd=3426, url=null, language=null, rfNumber=null, rfOrder=87, authorNames=Zhang D X, Yu J H, Zhang C Y, Cai W D, journalName=null, refType=null, unstructuredReference=Zhang D XYu J HZhang C Y and Cai W D. 2023. PaRot: patch-wise rotation-invariant network via feature disentanglement and pose restoration//Proceedings of the 37th AAAI Conference on Artificial Intelligence. Washington, USA: AAAI:3418-3426 [DOI: 10.1609/aaai.v37i3.25450], articleTitle=PaRot: patch-wise rotation-invariant network via feature disentanglement and pose restoration, refAbstract=null), Reference(id=1249044044756750848, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024a, volume=26, issue=null, pageStart=8052, pageEnd=8062, url=null, language=null, rfNumber=null, rfOrder=88, authorNames=Zhang H, Wang C S, Yu L, Tian S W, Ning X, Rodrigues J, journalName=IEEE Transactions on Multimedia, refType=null, unstructuredReference=Zhang HWang C SYu LTian S WNing X and Rodrigues J. 2024a. PointGT: a method for point-cloud classification and segmentation based on local geometric transformation. IEEE Transactions on Multimedia26: 8052-8062 [DOI: 10.1109/TMM.2024.3374580], articleTitle=PointGT: a method for point-cloud classification and segmentation based on local geometric transformation, refAbstract=null), Reference(id=1249044044823859713, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020a, volume=null, issue=null, pageStart=200, pageEnd=209, url=null, language=null, rfNumber=null, rfOrder=89, authorNames=Zhang J M, Yu M Y, Vasudevan R and Johnson-Roberson M, journalName=null, refType=null, unstructuredReference=Zhang J MYu M YVasudevan R and Johnson-Roberson M. 2020a. Learning rotation-invariant representations of point clouds using aligned edge convolutional neural networks//Proceedings of 2020 International Conference on 3D Vision. Fukuoka, Japan: IEEE:200-209 [DOI: 10.1109/3DV50981.2020.00030], articleTitle=Learning rotation-invariant representations of point clouds using aligned edge convolutional neural networks, refAbstract=null), Reference(id=1249044044916134402, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019a, volume=6, issue=1, pageStart=null, pageEnd=11, url=null, language=null, rfNumber=null, rfOrder=90, authorNames=Zhang S, Tong H H, Xu J J, Maciejewski R, journalName=Computational Social Networks, refType=null, unstructuredReference=Zhang STong H HXu J J and Maciejewski R. 2019a. Graph convolutional networks: a comprehensive review. Computational Social Networks6(1): #11 [DOI: 10.1186/s40649-019-0069-y], articleTitle=Graph convolutional networks: a comprehensive review, refAbstract=null), Reference(id=1249044044995826179, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020b, volume=null, issue=null, pageStart=7302, pageEnd=7311, url=null, language=null, rfNumber=null, rfOrder=91, authorNames=Zhang X, Qin S F, Xu Y, Xu H T, journalName=null, refType=null, unstructuredReference=Zhang XQin S FXu Y and Xu H T. 2020b. Quaternion product units for deep learning on 3D rotation groups//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE:7302-7311 [DOI: 10.1109/CVPR42600.2020.00733], articleTitle=Quaternion product units for deep learning on 3D rotation groups, refAbstract=null), Reference(id=1249044045075517956, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2024b, volume=212, issue=null, pageStart=58, pageEnd=72, url=null, language=null, rfNumber=null, rfOrder=92, authorNames=Zhang Z H, Liu H, Wang X, Shu M C, Chen G L, Zhang Q Z, journalName=ISPRS Journal of Photogrammetry and Remote Sensing, refType=null, unstructuredReference=Zhang Z HLiu HWang XShu M CChen G L and Zhang Q Z. 2024b. Lightweight and rotation-invariant place recognition network for large-scale raw point clouds. ISPRS Journal of Photogrammetry and Remote Sensing212: 58-72 [DOI: 10.1016/j.isprsjprs.2024.04.030], articleTitle=Lightweight and rotation-invariant place recognition network for large-scale raw point clouds, refAbstract=null), Reference(id=1249044045167792645, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=25, issue=8, pageStart=1551, pageEnd=1557, url=null, language=null, rfNumber=null, rfOrder=93, authorNames=Zhang X L, Fu C L, Zhang Y J, journalName=Journal of Image and Graphics, refType=null, unstructuredReference=Zhang X LFu C L and Zhang Y J. 2020. Extended pointwise convolution network model for point cloud classification and segmentation. Journal of Image and Graphics25(8): 1551-1557, articleTitle=Extended pointwise convolution network model for point cloud classification and segmentation, refAbstract=null), Reference(id=1249044045239095814, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=25, issue=8, pageStart=1551, pageEnd=1557, url=null, language=null, rfNumber=null, rfOrder=94, authorNames=张新良, 付陈琳, 赵运基, journalName=中国图象图形学报, refType=null, unstructuredReference=张新良, 付陈琳, 赵运基. 2020. 扩展点态卷积网络的点云分类分割模型. 中国图象图形学报25(8): 1551-1557 [DOI: 10.11834/jig.190508], articleTitle=扩展点态卷积网络的点云分类分割模型, refAbstract=null), Reference(id=1249044045314593287, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020c, volume=null, issue=null, pageStart=210, pageEnd=219, url=null, language=null, rfNumber=null, rfOrder=95, authorNames=Zhang Z Y, Hua B S, Chen W, Tian Y B, Yeung S K, journalName=null, refType=null, unstructuredReference=Zhang Z YHua B SChen WTian Y B and Yeung S K. 2020c. Global context aware convolutions for 3D point cloud understanding//Proceedings of 2020 International Conference on 3D Vision. Fukuoka, Japan: IEEE:210-219 [DOI: 10.1109/3DV50981.2020.00031], articleTitle=Global context aware convolutions for 3D point cloud understanding, refAbstract=null), Reference(id=1249044045402673672, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2019b, volume=null, issue=null, pageStart=204, pageEnd=213, url=null, language=null, rfNumber=null, rfOrder=96, authorNames=Zhang Z Y, Hua B S, Rosen D W, Yeung S K, journalName=null, refType=null, unstructuredReference=Zhang Z YHua B SRosen D W and Yeung S K. 2019b. Rotation invariant convolutions for 3D point clouds deep learning//Proceedings of 2019 International Conference on 3D Vision. Quebec City, Canada: IEEE:204-213 [DOI: 10.1109/3DV.2019.00031], articleTitle=Rotation invariant convolutions for 3D point clouds deep learning, refAbstract=null), Reference(id=1249044045536891401, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=130, issue=5, pageStart=1228, pageEnd=1243, url=null, language=null, rfNumber=null, rfOrder=97, authorNames=Zhang Z Y, Hua B S, Yeung S K, journalName=International Journal of Computer Vision, refType=null, unstructuredReference=Zhang Z YHua B S and Yeung S K. 2022. RIConv++: effective rotation invariant convolutions for 3D point clouds deep learning. International Journal of Computer Vision130(5): 1228-1243 [DOI: 10.1007/s11263-022-01601-z], articleTitle=RIConv++: effective rotation invariant convolutions for 3D point clouds deep learning, refAbstract=null), Reference(id=1249044045687886346, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2025, volume=null, issue=null, pageStart=93, pageEnd=109, url=null, language=null, rfNumber=null, rfOrder=98, authorNames=Zhang Z Y, Yang L C, Xiang Z Y, journalName=null, refType=null, unstructuredReference=Zhang Z YYang L C and Xiang Z Y. 2025. RISurConv: rotation invariant surface attention-augmented convolutions for 3D point cloud classification and segmentation//Proceedings of the 18th European Conference on Computer Vision. Milan, Italy: Springer:93-109 [DOI: 10.1007/978-3-031-73390-1_6], articleTitle=RISurConv: rotation invariant surface attention-augmented convolutions for 3D point cloud classification and segmentation, refAbstract=null), Reference(id=1249044045805326859, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2022, volume=127, issue=null, pageStart=null, pageEnd=108626, url=null, language=null, rfNumber=null, rfOrder=99, authorNames=Zhao C, Yang J Q, Xiong X, Zhu A F, Cao Z G, Li X, journalName=Pattern Recognition, refType=null, unstructuredReference=Zhao CYang J QXiong XZhu A FCao Z G and Li X. 2022. Rotation invariant point cloud analysis: where local geometry meets global topology. Pattern Recognition127: #108626 [DOI: 10.1016/j.patcog.2022.108626], articleTitle=Rotation invariant point cloud analysis: where local geometry meets global topology, refAbstract=null), Reference(id=1249044045872435724, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2021, volume=null, issue=null, pageStart=16239, pageEnd=16248, url=null, language=null, rfNumber=null, rfOrder=100, authorNames=Zhao H S, Jiang L, Jia J Y, Torr P, Koltun V, journalName=null, refType=null, unstructuredReference=Zhao H SJiang LJia J YTorr P and Koltun V. 2021. Point transformer//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:16239-16248 [DOI: 10.1109/ICCV48922.2021.01595], articleTitle=Point transformer, refAbstract=null), Reference(id=1249044045935350285, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2020, volume=null, issue=null, pageStart=1, pageEnd=19, url=null, language=null, rfNumber=null, rfOrder=101, authorNames=Zhao Y H, Birdal T, Lenssen J E, Menegatti E, Guibas L, Tombari F, journalName=Proceedings of the 16th European conference on computer vision, refType=null, unstructuredReference=Zhao Y HBirdal TLenssen J EMenegatti EGuibas L and Tombari F. 2020. Quaternion equivariant capsule networks for 3D point clouds//Proceedings of the 16th European conference on computer vision. Glasgow, UK: Springer:1-19 [DOI: 10.1007/978-3-030-58452-8_1], articleTitle=Quaternion equivariant capsule networks for 3D point clouds, refAbstract=null), Reference(id=1249044046015042062, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, doi=null, pmid=null, pmcid=null, year=2023, volume=null, issue=null, pageStart=1223, pageEnd=1232, url=null, language=null, rfNumber=null, rfOrder=102, authorNames=Zhu M H, Ghaffari M, Clark W A, Peng H E, journalName=null, refType=null, unstructuredReference=Zhu M HGhaffari MClark W A and Peng H E. 2023. E2PN: efficient SE(3)-equivariant point network//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver, Canada: IEEE:1223-1232 [DOI: 10.1109/CVPR52729.2023.00124], articleTitle=E2PN: efficient SE(3)-equivariant point network, refAbstract=null)], funds=null, companyList=[AuthorCompany(id=1249044020605947922, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, xref=1, ext=[AuthorCompanyExt(id=1249044020614336531, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China), AuthorCompanyExt(id=1249044020622725140, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, companyId=1249044020605947922, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1西北工业大学计算机学院,西安710072)])], figs=[ArticleFig(id=1249044028277330129, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.1, caption=Taxonomy of rotation invariant point cloud networks, figureFileSmall=DWmwQfb4D8afapsiGkH9HA==, figureFileBig=D2HeI8qhdFT5bRknRj6KaA==, tableContent=null), ArticleFig(id=1249044028365410518, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图1, caption=旋转不变点云网络总体分类图, figureFileSmall=DWmwQfb4D8afapsiGkH9HA==, figureFileBig=D2HeI8qhdFT5bRknRj6KaA==, tableContent=null), ArticleFig(id=1249044028591902944, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.2, caption=Chronological overview of representative rotation invariant point cloud networks, figureFileSmall=5JK0Kc+UjYbwWX5I1fn9/w==, figureFileBig=6AbAPMd5PIncKguqif/LjA==, tableContent=null), ArticleFig(id=1249044028696760548, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图2, caption=旋转不变点云网络典型方法发展脉络图, figureFileSmall=5JK0Kc+UjYbwWX5I1fn9/w==, figureFileBig=6AbAPMd5PIncKguqif/LjA==, tableContent=null), ArticleFig(id=1249044028789035238, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.3, caption=Overall framework of this review, figureFileSmall=xwdKsmZSkWJ02xdR1jdYrg==, figureFileBig=V4Tnj0bRS9/cdFM8uqb+8w==, tableContent=null), ArticleFig(id=1249044028872921323, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图3, caption=全文结构框架图, figureFileSmall=xwdKsmZSkWJ02xdR1jdYrg==, figureFileBig=V4Tnj0bRS9/cdFM8uqb+8w==, tableContent=null), ArticleFig(id=1249044028961001713, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.4, caption=The logical framework of geometric-guided rotation invariant point cloud networks, figureFileSmall=amYg5MT4xcX5e4IJCtSSKQ==, figureFileBig=d/DbyO19lhurpdIYOy7ZHg==, tableContent=null), ArticleFig(id=1249044029044887798, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图4, caption=几何旋转不变方法逻辑示意图, figureFileSmall=amYg5MT4xcX5e4IJCtSSKQ==, figureFileBig=d/DbyO19lhurpdIYOy7ZHg==, tableContent=null), ArticleFig(id=1249044029124579579, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.5, caption=The illustration of RIConv operator construction, figureFileSmall=W5e/rtzC85L+CKtw3LPT+w==, figureFileBig=vwzYRWjDqgli3ndIJu1slQ==, tableContent=null), ArticleFig(id=1249044029212659968, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图5, caption=RIConv卷积算子, figureFileSmall=W5e/rtzC85L+CKtw3LPT+w==, figureFileBig=vwzYRWjDqgli3ndIJu1slQ==, tableContent=null), ArticleFig(id=1249044029304934661, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.6, caption=The ambiguity of similar geometry in different local regions, figureFileSmall=2W/6OPyN6gGXTDqltdfpHQ==, figureFileBig=gQ00MAHZUUumKn5Zroy2Yw==, tableContent=null), ArticleFig(id=1249044029397209354, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图6, caption=不同局部邻域几何属性高度相似致使模糊性, figureFileSmall=2W/6OPyN6gGXTDqltdfpHQ==, figureFileBig=gQ00MAHZUUumKn5Zroy2Yw==, tableContent=null), ArticleFig(id=1249044029514649872, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.7, caption=The illustration of RI-GCN’s local feature extraction module, figureFileSmall=uMImjkT/AMz9aixB4ord5g==, figureFileBig=4FBaugZoySxTOEbIC01cbw==, tableContent=null), ArticleFig(id=1249044031091708180, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图7, caption=RI-GCN局部特征提取流程, figureFileSmall=uMImjkT/AMz9aixB4ord5g==, figureFileBig=4FBaugZoySxTOEbIC01cbw==, tableContent=null), ArticleFig(id=1249044031200760090, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.8, caption=The logical framework of feature-guided rotation invariant point cloud networks, figureFileSmall=KgdM2QsqlTKPNdOfRNi4JQ==, figureFileBig=jn2t0gW/dqipWr39EFvhYQ==, tableContent=null), ArticleFig(id=1249044031347560733, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图8, caption=特征旋转不变方法逻辑示意图, figureFileSmall=KgdM2QsqlTKPNdOfRNi4JQ==, figureFileBig=jn2t0gW/dqipWr39EFvhYQ==, tableContent=null), ArticleFig(id=1249044031452418338, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.9, caption=The overall architecture of Rot-SO-Net, figureFileSmall=SgcNsjXouJPPmFtn80mF+g==, figureFileBig=7nPU1gxUGqe1O4JD28xMiA==, tableContent=null), ArticleFig(id=1249044031527915814, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图9, caption=Rot-SO-Net网络框架, figureFileSmall=SgcNsjXouJPPmFtn80mF+g==, figureFileBig=7nPU1gxUGqe1O4JD28xMiA==, tableContent=null), ArticleFig(id=1249044031624384811, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.10, caption=The overall architecture of PRIN, figureFileSmall=yg/SX8S/fsVfRfHV7cLo0g==, figureFileBig=byHZWGi8mJV5XjGuWDM4Rg==, tableContent=null), ArticleFig(id=1249044031733436722, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图10, caption=PRIN网络框架, figureFileSmall=yg/SX8S/fsVfRfHV7cLo0g==, figureFileBig=byHZWGi8mJV5XjGuWDM4Rg==, tableContent=null), ArticleFig(id=1249044031800545588, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.11, caption=Comparison of scalar neurons and vector neurons, figureFileSmall=O5reJGbYlPnqnC2G5RbDZw==, figureFileBig=uWWxbxblt8G7ih4pSwNXWQ==, tableContent=null), ArticleFig(id=1249044031876043062, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图11, caption=标量神经元与向量神经元对比图, figureFileSmall=O5reJGbYlPnqnC2G5RbDZw==, figureFileBig=uWWxbxblt8G7ih4pSwNXWQ==, tableContent=null), ArticleFig(id=1249044031951540540, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.12, caption=The illustration of the rotation-equivariance of REQNN’s quaternion features, figureFileSmall=1coWYXdDRQaCekG0Hqe7Tw==, figureFileBig=Y7sf+DEMmhjtsrfdo2ZX4Q==, tableContent=null), ArticleFig(id=1249044032043815231, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图12, caption=REQNN四元数旋转等变特征, figureFileSmall=1coWYXdDRQaCekG0Hqe7Tw==, figureFileBig=Y7sf+DEMmhjtsrfdo2ZX4Q==, tableContent=null), ArticleFig(id=1249044032136089921, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Fig.13, caption=The illustration of the T-Net module, figureFileSmall=ZPsFSfoCk3gs5j7NLxttzA==, figureFileBig=h6oTQbubIVTgYx9BdOwpCw==, tableContent=null), ArticleFig(id=1249044032245141830, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=图13, caption=T-Net模块, figureFileSmall=ZPsFSfoCk3gs5j7NLxttzA==, figureFileBig=h6oTQbubIVTgYx9BdOwpCw==, tableContent=null), ArticleFig(id=1249044032349999433, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Tab.1, caption=

Classification performance of representative rotation invariant point cloud networks in the ModelNet40 dataset

, figureFileSmall=null, figureFileBig=null, tableContent=
类别方法总体分类精度(OA)
Z/ZZ/SO(3)SO(3)/SO(3)
几何旋转不变方法ClusterNet(Chen等,2019)87.187.187.1
RIConv(Zhang等,2019b)86.586.486.4
RI-GCN(Kim等,2020)89.589.589.5
PRinvNet(Yu等,2020)89.289.289.2
SGMNet(Xu等,2021)90.090.090.0
RI-framework(Li等,2022)89.489.489.3
LGRNet(Zhao等,2022)90.991.190.9
PARI-Conv(Chen和Cong,2022)91.491.491.4
CRIN(Lou等,2023)91.891.891.8
RISurConv(Zhang等,2025)95.695.695.6
LocoTrans(Chen等,2024)91.691.691.5
EIP*(Fei和Deng,2024a)-89.6-
GEConvNet(Bello等,2025)91.791.791.7
GLC-HCAN(Dai等,2025)92.592.592.5
RotInv-PCT(He等,2025)91.191.191.1
特征旋转不变方法TFN(Thomas等,2018)88.585.387.6
PRIN*(You等,2020)-72.4-
REQNN*(Shen等,2020)-84.6-
QEC-Net*(Zhao等,2020)-74.1-
SPRIN*(You等,2022)-86.1-
VN-DGCNN(Deng等,2021a)89.589.590.2
E2PN*(Zhu等,2023)90.544.488.6
TetraSphere(Melnyk等,2024)90.590.5
SE3Conv3D(Weijler和Hermosilla,2025)-87.089.0
训练旋转不变方法RTN(Deng等,2021b)--90.2
ART-Point(Wang等,2022)--90.5
SPE-Net(Zhang等,2022)92.789.791.8
PaRot(Zhang等,2023)90.991.090.8
), ArticleFig(id=1249044032438079821, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=表1, caption=

ModelNet40数据集下各类旋转不变点云网络代表性方法的分类性能

, figureFileSmall=null, figureFileBig=null, tableContent=
类别方法总体分类精度(OA)
Z/ZZ/SO(3)SO(3)/SO(3)
几何旋转不变方法ClusterNet(Chen等,2019)87.187.187.1
RIConv(Zhang等,2019b)86.586.486.4
RI-GCN(Kim等,2020)89.589.589.5
PRinvNet(Yu等,2020)89.289.289.2
SGMNet(Xu等,2021)90.090.090.0
RI-framework(Li等,2022)89.489.489.3
LGRNet(Zhao等,2022)90.991.190.9
PARI-Conv(Chen和Cong,2022)91.491.491.4
CRIN(Lou等,2023)91.891.891.8
RISurConv(Zhang等,2025)95.695.695.6
LocoTrans(Chen等,2024)91.691.691.5
EIP*(Fei和Deng,2024a)-89.6-
GEConvNet(Bello等,2025)91.791.791.7
GLC-HCAN(Dai等,2025)92.592.592.5
RotInv-PCT(He等,2025)91.191.191.1
特征旋转不变方法TFN(Thomas等,2018)88.585.387.6
PRIN*(You等,2020)-72.4-
REQNN*(Shen等,2020)-84.6-
QEC-Net*(Zhao等,2020)-74.1-
SPRIN*(You等,2022)-86.1-
VN-DGCNN(Deng等,2021a)89.589.590.2
E2PN*(Zhu等,2023)90.544.488.6
TetraSphere(Melnyk等,2024)90.590.5
SE3Conv3D(Weijler和Hermosilla,2025)-87.089.0
训练旋转不变方法RTN(Deng等,2021b)--90.2
ART-Point(Wang等,2022)--90.5
SPE-Net(Zhang等,2022)92.789.791.8
PaRot(Zhang等,2023)90.991.090.8
), ArticleFig(id=1249044032517771600, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Tab.2, caption=

ShapeNetPart segmentation performance of representative rotation invariant point cloud networks

, figureFileSmall=null, figureFileBig=null, tableContent=
类别方法年份实例均值IoU (ins.)类均值 IoU (cls)
Z/ZZ/SO(3)SO(3)/SO(3)Z/ZZ/SO(3)SO(3)/SO(3)
几何旋转不变方法RIConv(Zhang等,2019b)2019-80.280.2-75.375.5
RI-GCN(Kim等,2020)2020----77.277.3
PRinvNet(Yu等,2020)2020---79.479.479.4
SGMNet(Xu等,2021)2021---79.379.379.3
RI-framework(Li等,2022)2022-82.082.3-79.279.4
LGRNet(Zhao等,2022)2022-82.482.7-80.080.1
PARI-Conv(Chen和Cong,2022)202283.883.8---
CRIN(Lou等,2023)2023---80.580.580.5
LocoTrans(Chen等,2024)2024-84.083.8-80.180.0
EIP*(Fei和Deng,2024a)2024-84.9--82.1-
RISurConv(Zhang等,2025)2025----81.381.3
GEConvNet(Bello等,2025)2025--82.582.5
RotInv-PCT(He等,2025)2025---82.382.382.3
特征旋转不变方法PRIN*(You等,2020)2020-71.2--66.8-
SPRIN*(You等,2022)2021-82.7--79.5-
VN-DGCNN(Deng等,2021a)2021-81.481.4---
TetraSphere(Melnyk等,2024)202482.382.3----
训练旋转不变方法RTN(Deng等,2021b)2021--82.8---
SPE-Net(Zhang等,2022)2022-87.187.8---
PaRot(Zhang等,2023)2023----79.279.5
), ArticleFig(id=1249044032614240595, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=表2, caption=

ShapeNetPart数据集下各类旋转不变点云网络代表性方法的分割性能

, figureFileSmall=null, figureFileBig=null, tableContent=
类别方法年份实例均值IoU (ins.)类均值 IoU (cls)
Z/ZZ/SO(3)SO(3)/SO(3)Z/ZZ/SO(3)SO(3)/SO(3)
几何旋转不变方法RIConv(Zhang等,2019b)2019-80.280.2-75.375.5
RI-GCN(Kim等,2020)2020----77.277.3
PRinvNet(Yu等,2020)2020---79.479.479.4
SGMNet(Xu等,2021)2021---79.379.379.3
RI-framework(Li等,2022)2022-82.082.3-79.279.4
LGRNet(Zhao等,2022)2022-82.482.7-80.080.1
PARI-Conv(Chen和Cong,2022)202283.883.8---
CRIN(Lou等,2023)2023---80.580.580.5
LocoTrans(Chen等,2024)2024-84.083.8-80.180.0
EIP*(Fei和Deng,2024a)2024-84.9--82.1-
RISurConv(Zhang等,2025)2025----81.381.3
GEConvNet(Bello等,2025)2025--82.582.5
RotInv-PCT(He等,2025)2025---82.382.382.3
特征旋转不变方法PRIN*(You等,2020)2020-71.2--66.8-
SPRIN*(You等,2022)2021-82.7--79.5-
VN-DGCNN(Deng等,2021a)2021-81.481.4---
TetraSphere(Melnyk等,2024)202482.382.3----
训练旋转不变方法RTN(Deng等,2021b)2021--82.8---
SPE-Net(Zhang等,2022)2022-87.187.8---
PaRot(Zhang等,2023)2023----79.279.5
), ArticleFig(id=1249044032723292502, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=EN, label=Tab.3, caption=

Summary of different types of rotation invariant point cloud networks

, figureFileSmall=null, figureFileBig=null, tableContent=
方法类别代表性算法特点优势劣势
几何旋转不变

RIConv(Zhang等,2019b)

LGRNet(Zhao等,2022)

PARI-Conv(Chen和Cong,2022b)

EIP(Fei和Deng,2024a)

RISurConv(Zhang等,2025)

基于传统几何分析(距离、夹角、局部参考系、规范姿态等),在旋转不变的几何属性空间或坐标空间上提取点云特征。理论较为简单,可解释性强。相关点云网络通过几何预处理消除旋转影响,直接保证旋转不变性。相较于其他两类方法,传统几何分析预处理存在一定的信息损失。
特征旋转不变

TFN(Thomas等,2018)

PRIN(You等,2020)

REQNN(Shen等,2020)

TetraSphere(Melnyk等,2024)

SE3Conv3D

(Weijler和Hermosilla,2025)

利用旋转等变点云网络提取旋转等变特征,通过后处理(池化、内积等)剥离等变特征中的姿态信息,获得旋转不变点云特征。理论完备性高。相关点云网络直接输入点云数据,接收信息更为完整。复杂的网络设计理论使得相关方法难以拓展到更大规模的点云数据,并且后处理步骤难以保证完整姿态无关信息的保留。
训练旋转不变

RTN(Deng等,2021b)

ART-Point(Wang等,2022)

SPE-Net(Zhang等,2022)

PaRot(Zhang等,2023)

通过数据增强(随机旋转)、添加子模块(T-Net)以及复杂训练策略(对抗训练、孪生学习)驱动常规点云网络学习旋转不变性约束。易与前沿非旋转不变点云网络相结合,无须修改点云网络基本结构。无法严格保证旋转不变性,且SO(3)群的无限连续性使旋转增强难以覆盖所有姿态,目前缺乏强有力的理论支持。
), ArticleFig(id=1249044032815567195, tenantId=1146029695717560320, journalId=1249024232475115590, articleId=1249044010443150108, language=CN, label=表3, caption=

各类旋转不变点云网络定性归纳总结

, figureFileSmall=null, figureFileBig=null, tableContent=
方法类别代表性算法特点优势劣势
几何旋转不变

RIConv(Zhang等,2019b)

LGRNet(Zhao等,2022)

PARI-Conv(Chen和Cong,2022b)

EIP(Fei和Deng,2024a)

RISurConv(Zhang等,2025)

基于传统几何分析(距离、夹角、局部参考系、规范姿态等),在旋转不变的几何属性空间或坐标空间上提取点云特征。理论较为简单,可解释性强。相关点云网络通过几何预处理消除旋转影响,直接保证旋转不变性。相较于其他两类方法,传统几何分析预处理存在一定的信息损失。
特征旋转不变

TFN(Thomas等,2018)

PRIN(You等,2020)

REQNN(Shen等,2020)

TetraSphere(Melnyk等,2024)

SE3Conv3D

(Weijler和Hermosilla,2025)

利用旋转等变点云网络提取旋转等变特征,通过后处理(池化、内积等)剥离等变特征中的姿态信息,获得旋转不变点云特征。理论完备性高。相关点云网络直接输入点云数据,接收信息更为完整。复杂的网络设计理论使得相关方法难以拓展到更大规模的点云数据,并且后处理步骤难以保证完整姿态无关信息的保留。
训练旋转不变

RTN(Deng等,2021b)

ART-Point(Wang等,2022)

SPE-Net(Zhang等,2022)

PaRot(Zhang等,2023)

通过数据增强(随机旋转)、添加子模块(T-Net)以及复杂训练策略(对抗训练、孪生学习)驱动常规点云网络学习旋转不变性约束。易与前沿非旋转不变点云网络相结合,无须修改点云网络基本结构。无法严格保证旋转不变性,且SO(3)群的无限连续性使旋转增强难以覆盖所有姿态,目前缺乏强有力的理论支持。
)], attaches=null, journal=Journal(id=1249023527618129992, delFlag=0, nameCn=中国图象图形学报, nameEn=Journal of Image and Graphics, nameHistory1=null, nameHistory2=null, issn=1006-8961, eissn=null, cn=11-3758, coden=CODEN ZTTXFZ, 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=uirXtX858YS3zEpFXZttJA==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Image and Graphics, journalRemark=null, publicationField=null, createdTime=1775720014721, updatedTime=1775720337198, createdBy=18614031015, updatedBy=13701087609, firstLetterCn=J, firstLetterEn=J, subjectCode=Engineering, subjectName=null, subjectCodeEn=Engineering, subjectNameEn=null, picCn=uirXtX858YS3zEpFXZttJA==, picEn=bud7qaxfvWHeFsbyBTAiKQ==, jcr=null, cjcr=null, exts=[JournalExt(id=1249024880377786590, language=CN, name=中国图象图形学报, 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=1775720337242, updatedTime=1775720337242, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-author&redirect_uri=https%3A%2F%2Fcjig.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=e6369def-2842-41d8, submissionEditorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-editor&redirect_uri=https%3A%2F%2Fcjigeditor.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=9ccec05b-6bd, submissionReviewUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-author&redirect_uri=https%3A%2F%2Fcjig.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=1e8a31c8-5434-4f78, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1249024880449089759, language=EN, name=Journal of Image and Graphics, 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=1775720337259, updatedTime=1775720337259, createdBy=13701087609, updatedBy=13701087609, submissionGuidelinesUrl=, submissionAuthorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-author&redirect_uri=https%3A%2F%2Fcjig.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=e6369def-2842-41d8, submissionEditorUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-editor&redirect_uri=https%3A%2F%2Fcjigeditor.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=9ccec05b-6bd, submissionReviewUrl=https://journal.ids.fzyun.cn/auth/realms/journal/protocol/openid-connect/auth?client_id=journal-cjig-author&redirect_uri=https%3A%2F%2Fcjig.portal.founderss.cn%2Foauth%2Fcallback&response_type=code&scope=phone+openid+email+profile&state=1e8a31c8-5434-4f78, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1249024232475115590, websiteList=[Website(id=1249025782459334881, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1249024232475115590, 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/zgtxtxxb/CN, language=CN, createTime=1775720552315, createBy=18614031015, updateTime=1775720586268, updateBy=18614031015, name=中国图象图形学报-中文, tplId=1146099689490845704, title=中国图象图形学报, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1249026166254928133, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=articleTextType, value=kx, createTime=1775720643819, updateTime=1775720643819, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166221373698, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=banner, value=null, createTime=1775720643811, updateTime=1775720643811, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166271705352, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=grayFlag, value=0, createTime=1775720643823, updateTime=1775720643823, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166212985089, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=logo, value=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/file/pic?fileId=TDRjKTHfgAnvFKZaDA70wA==, createTime=1775720643809, updateTime=1775720643809, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166288482570, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=minRunFlag, value=0, createTime=1775720643827, updateTime=1775720643827, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166246539524, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/file/pic, createTime=1775720643817, updateTime=1775720643817, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166280093961, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=silenceFlag, value=0, createTime=1775720643825, updateTime=1775720643825, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166233956611, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1775720643814, updateTime=1775720643814, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166259122438, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=themeColor, value=null, createTime=1775720643820, updateTime=1775720643820, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026166267511047, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782459334881, code=themeStyle, value=null, createTime=1775720643822, updateTime=1775720643822, creator=18614031015, updator=18614031015)]), Website(id=1249025782681633001, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1249024232475115590, 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/zgtxtxxb/EN, language=EN, createTime=1775720552368, createBy=18614031015, updateTime=1775720607118, updateBy=18614031015, name=中国图象图形学报-英文, tplId=1146101810881728533, title=Journal of Image and Graphics, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1249026195371786511, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=articleTextType, value=kx, createTime=1775720650761, updateTime=1775720650761, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195355009292, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=banner, value=null, createTime=1775720650757, updateTime=1775720650757, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195392758034, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=grayFlag, value=0, createTime=1775720650766, updateTime=1775720650766, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195342426379, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=logo, value=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN/file/pic?fileId=TDRjKTHfgAnvFKZaDA70wA==, createTime=1775720650754, updateTime=1775720650754, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195409535252, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=minRunFlag, value=0, createTime=1775720650770, updateTime=1775720650770, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195367592206, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN/file/pic, createTime=1775720650760, updateTime=1775720650760, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195401146643, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=silenceFlag, value=0, createTime=1775720650768, updateTime=1775720650768, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195359203597, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1775720650758, updateTime=1775720650758, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195380175120, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=themeColor, value=null, createTime=1775720650763, updateTime=1775720650763, creator=18614031015, updator=18614031015), WebsiteProps(id=1249026195388563729, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1249025782681633001, code=themeStyle, value=null, createTime=1775720650765, updateTime=1775720650765, creator=18614031015, updator=18614031015)])], journalTitle=中国图象图形学报, weixinUrl=null, journalUrl=https://www.cjig.cn/, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Image and Graphics, journalPhotoCn=uirXtX858YS3zEpFXZttJA==, journalPhotoEn=bud7qaxfvWHeFsbyBTAiKQ==, 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), detailUrlCn=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/10.11834/jig.250029, detailUrlEn=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN/10.11834/jig.250029, pdfUrlCn=https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/PDF/10.11834/jig.250029, pdfUrlEn=https://castjournals.cast.org.cn/joweb/zgtxtxxb/EN/PDF/10.11834/jig.250029, aliStartDate=null, aliEndDate=null, collectionFlag=false, citedCount=null, citedUrl=null, reference=null)
收藏切换
旋转不变点云网络研究进展
收藏切换
PDF下载
王正宝 , 曾振轩 , 欧阳轩 , 陈昊哲 , 李林杰 , 杨佳琪 *
中国图象图形学报 | 综述 2025,30(12): 3782-3803
收起
收藏切换
中国图象图形学报 | 综述 2025, 30(12): 3782-3803
旋转不变点云网络研究进展
全屏
王正宝 , 曾振轩 , 欧阳轩 , 陈昊哲 , 李林杰 , 杨佳琪*
作者信息
  • 1西北工业大学计算机学院,西安710072
  • 王正宝,男,硕士研究生,主要研究方向为点云表征。E-mail:

    杨佳琪,通信作者,男,长聘副教授,主要研究方向为计算资源、数据标签受限条件下的三维配准重建。E-mail:

    曾振轩,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

    欧阳轩,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

    陈昊哲,男,硕士研究生,主要研究方向为非刚性点云配准。E-mail:

    李林杰,男,硕士研究生,主要研究方向为刚性点云配准。E-mail:

通讯作者:

Recent progress in rotation-invariant point cloud networks
Zhengbao Wang , Zhenxuan Zeng , Xuan Ouyang , Haozhe Chen , Linjie Li , Jiaqi Yang*
Affiliations
  • 1School of Computer Science,Northwestern Polytechnical University,Xi’an710072,China
出版时间: 2025-12-16 doi: 10.11834/jig.250029
文章导航
收藏切换

点云深度学习网络取得显著进展,表现出的强大语义理解能力推动着整个三维视觉领域的进步。然而三维点云通常表现出复杂的变换对称性,其中旋转是一个具有挑战性且必要的主题。不同旋转变换下点云的语义信息一致,但空间坐标不同,这影响了常规点云深度学习网络在语义感知方面的稳定性,难以应用到任意姿态的现实场景。早期的研究主要采用旋转数据增强的方式,但由于旋转本身的无限性和连续性,这种简易方案并不能满足需求。因此,越来越多的学者着手研究具有旋转不变属性的点云深度学习网络,在网络设计层面排除旋转对于特征提取的影响。本文对于旋转不变点云网络相关研究进行充分调研,分析其中存在的挑战,并系统整理相关主流方法,依据旋转不变能力获取方式的不同,将其划分为几何旋转不变方法、特征旋转不变方法和训练旋转不变方法。本文详细描述了当前学术界在该问题上的研究内容和方法,总结和对比各类方法的优缺点,并对常用的一些数据集和评价指标进行整理总结。最后,本文调研和总结了旋转不变点云深度学习网络的下游应用前景,并对未来发展和研究热点进行展望。

三维视觉  /  深度学习  /  点云网络  /  旋转不变性  /  旋转等变性

In recent years, deep learning networks for point clouds have achieved remarkable advancements, with their robust semantic understanding capabilities propelling research across the entire field of three-dimensional (3D) computer vision. These advancements have enabled accurate and efficient processing of 3D data, supporting applications in autonomous driving, robotics, remote sensing and mapping, and augmented reality. However, 3D point clouds often exhibit complex transformation symmetries, with rotation being a particularly challenging yet critical factor. The spatial coordinates of point clouds, which are the fundamental input to point cloud networks, undergo substantial changes, resulting in feature output variations. However, the semantic information embedded within point clouds theoretically remains consistent under various rotational transformations. This spatial variability substantially impacts the stability and reliability of conventional point cloud deep learning networks in semantic perception tasks, such as recognition, classification, and segmentation, reducing their effectiveness in real-world scenarios characterized by arbitrary orientations and poses. Early studies primarily relied on rotational data augmentation to enhance the robustness of point cloud networks against rotational variations. While data augmentation can improve generalization to some extent, it falls short of addressing the fundamental issue posed by the infinite and continuous nature of the rotation group. Acknowledging these limitations, an increasing number of researchers have shifted their focus toward designing rotation-invariant point cloud deep learning networks, which aim to mitigate the impact of rotation on feature extraction at the architectural level. Therefore, researchers seek to achieve consistent semantic perception regardless of point cloud orientation, thereby enhancing the applicability of deep learning models in real-world scenarios where data can be encountered in arbitrary poses. This paper presents a comprehensive survey of the current state of research on rotation-invariant point cloud networks. The research background is first outlined to highlight the importance of rotation invariance in 3D vision tasks and the challenges posed by rotational symmetries in point cloud data. Then, a systematic categorization of the prevailing mainstream methods is investigated. Particularly, the rotation-invariant point cloud networks can be broadly classified into the following three categories: 1) geometric-guided rotation-invariant methods: Using the traditional geometric analysis algorithms, these methods extract rotation-invariant geometric representations such as relative distances, angles, local reference frames, and canonical poses. These representations are then integrated into point cloud networks, facilitating learning of high-level semantic features and maintaining robustness to rotational transformations simultaneously. 2) Feature-guided rotation-invariant methods: These methods employ rotation-equivariant point cloud networks to extract point cloud representations that contain shape and pose information. Leveraging the inherent principles of equivariant networks, they subsequently remove the pose information from the rotation-equivariant representations, obtaining rotation-invariant point cloud features. 3) Training-guided rotation-invariant methods: These methods focus on designing sophisticated and highly generalizable rotational data augmentation training schemes, allowing non-rotation-invariant point cloud networks to gradually acquire robustness of rotations and achieve stable performance simultaneously. An in-depth analysis of the core concepts and algorithmic improvements that support these methods is provided for each category. The current research content on this issue and methodologies within the academic community are outlined, and the advantages and disadvantages of each method are summarized and compared. Subsequently, a comprehensive overview of the prevalent downstream tasks in the research of rotation-invariant point cloud networks is presented. These tasks include point cloud classification, point cloud segmentation, and point cloud retrieval. For each of these tasks, an in-depth discussion of the commonly employed datasets and evaluation metrics, which are essential for assessing network performance, is provided. Additionally, the quantitative performance metrics of mainstream rotation-invariant point cloud networks applied to these tasks are summarized and analyzed, offering a comparative perspective on their efficacy and robustness under rotational variations. Afterward, the downstream application prospects of rotation-invariant point cloud deep learning networks, including point cloud self-supervised representation learning, end-to-end point cloud registration, and point cloud completion, are examined and summarized. Finally, an outlook on future developments and research hotspots is presented. In addition to the ongoing development of new rotation-invariant point cloud networks, three primary issues warrant further research: 1) discrimination of effective geometric attributes. Current approaches are limited by the design of geometric attribute extraction algorithms. An in-depth discussion and determination of the effectiveness of different rotation-invariant geometric attributes within deep learning frameworks could yield novel insights and foster the development of innovative strategies to advance this field. 2) Highly integratable rotation-invariant mechanism. On the one hand, existing non-rotation-invariant point cloud networks continue to demonstrate strong performance on aligned data. The challenge lies in incorporating rotation invariance into these networks in a straightforward manner degrading their original performance. This challenge remains a key research topic because seamless integration requires innovative architectural designs and methodological approaches. On the other hand, rotation-invariant point cloud networks should also exhibit simplicity and reusability, enabling their direct application to downstream tasks with minimal adaptation. 3) High computational efficiency in invariant feature extraction modules. Although many existing methods demonstrate commendable performance, they often incur substantial time and computational costs, making it challenging to efficiently process large-scale point cloud data. Therefore, designing more efficient rotation-invariant point cloud networks that maintain robust feature extraction capabilities while minimizing computational overhead is crucial. Addressing the aforementioned challenges will notably enhance the effectiveness and practicality of rotation-invariant point cloud deep learning networks, facilitating their widespread adoption in complex 3D environments. This survey aims to provide researchers in 3D computer vision with a foundational understanding of current methodologies, highlight key challenges, and suggest potential avenues for future research.

three-dimensional vision  /  deep learning  /  point cloud network  /  rotation invariance  /  rotation equivariance
王正宝, 曾振轩, 欧阳轩, 陈昊哲, 李林杰, 杨佳琪. 旋转不变点云网络研究进展. 中国图象图形学报, 2025 , 30 (12) : 3782 -3803 . DOI: 10.11834/jig.250029
Zhengbao Wang, Zhenxuan Zeng, Xuan Ouyang, Haozhe Chen, Linjie Li, Jiaqi Yang. Recent progress in rotation-invariant point cloud networks[J]. Journal of Image and Graphics, 2025 , 30 (12) : 3782 -3803 . DOI: 10.11834/jig.250029
随着传感器技术的不断进步,点云以其易获取、精确表达空间和结构信息的优势,逐渐成为三维视觉中一种重要的数据模态,广泛应用在智能机器人(Mahler等,2017)、自动驾驶(Yan等,2018)、遥感测绘(Yu等,2019)和高精度地图(Liang等,2018)等领域。近年,与计算机视觉、语音处理和自然语言处理等领域一样,深度学习技术也主导了三维视觉领域的发展。研究者提出很多经典的点云智能感知与理解模型。例如PointNet(Qi等,2017a)、DGCNN(dynamic graph convolutional neural network)(Wang等,2019)、PointTransformer(Zhao等,2021)、PointMLP(point multilayer perceptron)(Ma等,2022)等。相关网络在点云补全(Huang等,2020Yu等,2021)、点云上采样(Li等,2023Kwon等,2023)、点云分类(张新良 等,2020Zhang等,2024a)、点云配准(Aoki等,2019Qin等,2023)等多个下游任务上大放异彩,不仅性能优于传统算法,还在多个任务中表现出很强的通用性。
而在真实场景中,点云往往存在姿态多样性、噪声干扰等问题,这对点云深度学习模型的鲁棒性提出严峻挑战。例如,在自动驾驶场景中,车辆传感器采集的点云数据可能因车辆运动或传感器安装角度不同而呈现任意旋转姿态,这直接影响了模型的感知精度和稳定性。尽管以PointNet、DGCNN等为代表的点云网络在分类、分割和识别等任务中表现优异,但其性能严重依赖于输入点云具有规范姿态。当测试数据存在任意旋转时,这些网络在相关任务上的性能将显著下降(Zhang等,2020a)。这一局限性严重制约了点云深度学习网络在实际工程中的应用。因此,当前研究亟需具有旋转不变性(rotation invariance)的点云深度学习网络,即网络在下游任务上的性能不受输入点云姿态影响。
对于旋转不变点云网络F来说,在实数域RR中,若对具有NN个点的输入点云PRN×3P∈RN×3施加任意旋转RR∈ SO(3)(special orthogonal group in 3 dimensions)(三维旋转群),网络输出的特征应能够保持一致,即满足
F(RP)=F(P)
早期,研究者通常在训练阶段应用随机旋转采样来减轻测试时的姿态变化影响(Choy等,2019)。然而,三维旋转群具有连续性和无限性,随机旋转采样难以覆盖所有可能姿态,进而点云网络也无法保证输入点云经任意旋转后,其输出特征不变,在具有随机旋转设置的下游任务测试上效果较差。所以,旋转不变点云网络的研究不是简单的数据增强问题,而是涉及几何、代数和深度学习理论的交叉挑战。研究者既要保证所提出点云网络的旋转不变性具有理论上的完备性和可解释性,也需要确保该网络在下游任务中获得良好的性能。
针对上述旋转不变点云网络的定义与要求,研究者从网络架构设计层次出发,提出3类主流解决方案:1)几何旋转不变方法。这类方法结合传统的几何分析算法,令网络中的特征提取算子作用在旋转不变空间上;2)特征旋转不变方法。这类方法基于数理知识构建具有旋转等变性的点云网络,对输入点云提取旋转等变特征,从而在特征空间上将其转换为旋转不变特征;3)训练旋转不变方法。这类方法通过更加细致、优化的训练方案,让原本不具有旋转不变属性的点云网络产生对旋转噪声的强鲁棒性。
图1所示,前两类解决方案可进一步细分。相关典型方法发展脉络如图2所示。这些方法在理论创新性和实际有效性上各有侧重,但目前对其设计机理、发展脉络及关联性的系统性总结仍较少(Fei和Deng,2024b)。为帮助读者全面地了解现有旋转不变点云网络领域的发展概况,本文将依次介绍这3类方法,并对相关点云网络进行总结。
在传统的点云处理算法中,研究者常常寻找旋转不变的几何属性或构建旋转不变的参考系来制作点云描述符,用于识别、匹配等任务。例如FPFH(fast point feature histograms)描述符(Rusu等,2009)根据点坐标和表面法线建立局部参考系(local reference frame,LRF),在参考系内部提取若干角度信息表征局部表面;SHOT(signatures of histograms for local surface)描述符(Tombari等,2010)定义了一种特异且鲁棒的局部参考系来获取稳健且富有表现力的三维描述符;PPF(point pair feature)描述符(Drost等,2010)则使用点对之间的相对距离、夹角等旋转不变的几何属性描述点云中所包含的信息。这些描述符具有较为稳定的旋转不变性,可以有效地处理识别、配准等任务。
基于此,点云深度学习研究者希望借助传统几何分析的旋转不变能力,让点云网络在学习高级语义特征的同时,保持对于旋转的鲁棒性,该类方法总体逻辑如图4所示。根据所使用分析技术的不同,本文将这些工作大致划分为如下类别:1)基于几何属性的旋转不变网络;2)基于局部参考系的旋转不变网络;3)基于全局规范姿态的旋转不变网络。
基于几何属性的旋转不变点云网络通常使用点对之间的相对距离、夹角等旋转不变几何属性表征点云结构,构建几何属性向量代替点云坐标作为特征提取的计算对象,即在旋转不变几何属性空间,对点云进行高维特征提取。在整个发展历程中,研究者提出多样的几何属性计算策略,其中主要的区别在于夹角的选用、参考点的选取等。纵观所有主流方法,可根据发展脉络大致划分为3个阶段。
第1阶段方法将几何属性的提取作为预处理步骤,与特征提取网络分离,代表性方法有:PPFNet(Deng等,2018b)、PPF-FoldNet(Deng等,2018a)、ClusterNet(Chen等,2019;李冠彬 等,2022)等。其中,PPFNet和PPF-FoldNet主要用于点云配准中的局部特征描述,网络首先计算关键点对应局部区域的PPF(point pair feature)表示,输入到PointNet中提取特征,构建局部描述符;ClusterNet则为输入点云中每个点构建邻域连通图,并计算邻域点和该点的相对角度,形成3K+13K+1维的旋转不变几何属性向量,用做网络输入。在网络结构部分,ClusterNet一方面设计相应模块压缩各点几何属性向量,保证后续特征提取效率;另一方面使用几何层次聚类树引导网络提取和聚合旋转不变特征。第1阶段方法虽然具有旋转不变性,但网络仅能用于输出描述符或分类概率向量,描述局部结构的能力较弱。
为此,第2阶段方法则是将几何属性与点云深度学习中各类局部特征提取算子耦合,设计旋转不变的局部特征提取算子,以实现点云网络的旋转不变性。具体而言,这些方法获得特征提取所需的邻域点集后,用邻域点的旋转不变几何属性向量代替点云坐标参与计算。相比于第1阶段方法,这些方法在保证网络局部结构描述能力的同时,更加高效地注入旋转不变性,代表性方法有:RIConv(rotation invariant convolution)(Zhang等,2019b)、SGMNet(sorted Gram matrix network)(Xu等,2021)、RI-Conv++(Zhang等,2022)和GEConvNet(geometric edge convolution)(Bello等,2025)等。其中,RIConv首先将几何属性嵌入到局部特征提取算子中,该网络以邻域质心和关键点作为参考点,与邻域中各个点建立几何分析三角形,以提取距离、夹角等不变几何属性,为邻域点构建了4维的几何属性向量,并将该向量作为点云卷积算子的输入,如图5所示。GEConvNet则在RIConv基础上构造以旋转不变几何特征为边的边缘卷积算子,大幅提高网络性能。然而,RI-Conv的几何属性仅描述邻域点相对关键点的分布,可能会损失邻域点之间的结构信息。对此,SGMNet引入排序格拉姆矩阵(sorted Gram matrix,SGM)作为几何表征,结合向量内积表示距离和夹角,建模邻域中的所有点对关系。SGMNet将该几何表征嵌入到PointNet++网络(Qi等,2017b)中,在局部特征提取时与点云特征拼接,以实现网络的旋转不变性;RI-Conv++则在RIConv基础上引入次邻近点,代替邻域质心,构建各邻域点的几何分析三角形,从而缓解信息损失的问题。
尽管第2阶段方法相比于第1阶段方法取得了更好的下游任务性能,但从全局尺度考虑,存在这一情况:不同局部邻域中的旋转不变几何属性向量高度相似。例如,飞机点云的两侧机翼是高度对称的,邻域内距离、夹角等旋转不变几何属性高度相似,如图6所示。这使得提取到的局部特征具有模糊性,影响了全局形状信息的感知。因此,第3阶段方法专注于旋转不变的全局形状建模,以解决这一模糊性问题。DH-RINet(deep hierarchical rotation invariant network)(Lin等,2021)和RI-framework(Li等,2022)尝试为每个点构建全局表征向量,与局部几何属性向量一起参与到邻域特征提取中。其中,DH-RINet用有限带宽球谐能量描述符建模全局形状,以获取逐点的全局表征,RI-framework则基于几何分析获取逐点全局表征,例如点向量模等。但是,这种耦合的方式可能会影响到局部尺度的特征提取,使得两个网络对于下游任务的性能提升并不明显。之后,LGRNet(local-global-representation network)(Zhao等,2022)、RLGRNet(rethinking local-to-global representation network)(Wang等,2024b)、ELGANet(enhanced local-global alignment network)(Gu等,2022)等方法尝试设计显式融合方案,向旋转不变局部几何特征注入全局信息。具体而言,这些方法基于双分支架构,一个分支进行局部几何特征提取,与第2阶段方法一致,另一个分支进行全局形状特征提取,另外配合各种显式特征融合策略,以解决形状信息损失问题。其中,LGRNet、RLGRNet用最远点采样(farthest point sampling,FPS)算法降采样输入点云,获得全局的骨架信息,结合奇异值分解(singular value decomposition,SVD)建立全局的点云规范姿态,从而在规范姿态点云下进行全局形状特征提取;ELGANet则使用提取的局部几何特征去回归变换矩阵,将点云转换到规范姿态,提取全局形状特征。
虽然第3阶段方法总体上性能优于前两个阶段的方法,但并不代表全局形状建模是几何属性不变点云网络后续唯一的发展方向。近期,属于第2阶段方法的工作也呈现出强大的特征提取能力,在各自任务上体现出了优异的性能。例如,RoITr(rotation-invariant transformer)(Yu等,2023a)仅基于PPF表示建模局部几何属性,配合较为复杂的注意力机制,设计出旋转不变的点云Transformer(Vaswani等,2017),在点云配准上取得了任意旋转数据设置下的最佳配准回召率;RISurConv(Zhang等,2025)构建局部三角形表面来捕获更详细的表面结构,提取高表达性的旋转不变几何属性,并结合自注意力机制设计旋转不变局部特征提取算子,实现了目前旋转不变点云网络中点云分类任务下的最佳性能。
上述基于几何属性的旋转不变点云网络实现了完备的旋转不变性,但距离、夹角等几何属性只是点云中包含的低级几何特征,点云网络能否基于这些低级几何特征提取到类别、形状等高级语义一直以来是一个无法清晰解释的问题(Lou等,2023)。因此,一些研究者尝试将LRF引入点云网络中,在局部特征提取时将邻域点坐标变换到LRF下,只要在任意姿态下,算法对于输入点云估计的LRF一致,所提取的特征便具有旋转不变性,即在旋转不变坐标空间上提取点云特征。
早期,RI-GCN(rotation invariant graph convolutional network)(Kim等,2020)用主成分分析(principal component analysis,PCA)算法计算邻域点集的局部参考系,将参考系下的邻域点坐标作为输入提取局部特征,如图7所示。在网络结构方面,该方法结合图卷积神经网络(graph convolutional network,GCN)(Zhang等,2019a)对局部特征进行聚合,并逐层使用FPS算法降采样点云以扩大卷积感受野,最终获取全局的旋转不变特征。
RI-GCN作为基于局部参考系实现旋转不变点云网络的首次尝试,存在以下局限性:1)基于PCA分析的局部参考系算法易受到遮挡、噪点和采样误差等点云噪声的影响,无法在带噪点云上保持稳定、输出一致的参考系估计;2)不同邻域点集可能具有不同的局部参考系,用各自局部参考系规范邻域点集位姿后,邻域与邻域之间的相对姿态可能发生变化,造成点云全局形状信息的损失。
对于第1个问题,一些研究者尝试提出更加鲁棒的局部参考系估计方法,增强网络的稳定性:GCANet(Zhang等,2020c)提出全局信息加权的局部参考系,用降采样得到的稀疏点云计算局部特征提取时关键点的LRF,虽然一定程度上提高了参考系的稳定性,但降低了计算效率;CRIN(centrifugal rotation invariant network)(Lou等,2023)提出离心参考系(centrifugal reference frame,CRF),这种参考系仅依靠关键点坐标和法向量获得,降低局部参考系计算时耗的同时,在一定程度上规避了遮挡、缺失等噪声,具有更强的稳定性。另一些研究者则将局部参考系的估计也纳入网络可学习范围,与特征提取主干耦合在一起训练,例如MA-KPC(multi-alignment kernel point convolution)(Thomas,2020)首先用不同尺度的邻域估计多个局部参考系,加权聚合多个参考系下的局部特征,其次在每次提取旋转不变局部特征后,都用该特征为不同尺度的参考系各回归一个旋转矩阵,对PCA分析得到的不稳定局部参考系进行动态修正;orientedMP(Luo等,2022)和LocoTrans(Chen等,2024)则引入向量神经网络(vector neuron network,VNN)(Deng等,2021a),这种网络能够输出由三维向量堆叠而成的旋转等变特征矩阵。因此,orientedMP和LocoTrans方法用VNN网络回归逐点的局部参考系。具体而言,两个方法用VNN-DGCNN模型根据输入点云,回归逐点的2×32×3维旋转等变向量特征矩阵,定义为局部参考系的两个轴,而后用向量叉乘得到第3个轴,得到逐点的局部参考系;Rotation3d(Yu等,2023b)在建立局部参考系的同时,也建立全局规范姿态,为点云中各点提取两个尺度下的旋转不变特征,并引入点云配准概念,约束两个尺度下的旋转不变特征逐渐对齐,从而获取跨尺度共享的语义信息,缓解参考系不稳定造成的几何特征抗旋转能力不鲁棒问题。
对于第2个问题,MA-KPC尝试在特征提取模块中计算邻域点LRF和关键点LRF的差异信息编码,与邻域点特征拼接,输入到特征提取模块进行计算。这种方式比较简便,单纯依靠网络对于所提供信息的识别能力;AECNN(aligned edge convolutional neural network)(Zhang等,2020a)则认为相对姿态问题出现在网络中更高层次的特征提取阶段,在这些阶段中,邻域点特征来自前一特征提取阶段的不同局部参考系,而网络没有将这些特征对齐就直接进行后续的计算、聚合步骤。因此,AECNN设计了一个特征对齐模块,输入邻域点特征、局部参考系的相对变换矩阵,输出估计的对齐邻域点特征,而后再进行本阶段的特征提取;基于AECNN的观点,PARI-Conv(pose-aware rotation invariant convolution)(Chen和Cong,2022)和LocoTrans方法尝试设计动态卷积算子,在特征提取时隐式地恢复相对姿态信息。其中,PARI-Conv基于邻域点LRF和关键点LRF之间的各轴夹角、邻域点和关键点之间的相对距离构建几何属性向量APPF(augmented point pair feature),并用其生成点云卷积核的权重,LocoTrans则利用VNN网络输出的等变特征矩阵编码每个邻居点与关键点之间的原始相对位姿,并利用该编码生成卷积核权重。RotInv-PCT(rotation invariant point cloud Transformer)(He等,2025)则直接编码LRF,并使用注意力机制补全特征中点云的全局形状信息。
无论是几何属性还是局部参考系点云网络,其旋转不变能力的获得均是以损失一定全局形状信息为代价的,这使得相关方法花费大量精力在恢复损失的信息中。因此,还有一些工作专注于寻找全局的唯一规范姿态,将该姿态下的点云特征作为旋转不变特征,从而保证全局形状信息不受损失,代表性方法有PRinvNet(Yu等,2020)和PCA-RINet(Li等,2021)。
其中,PRinvNet将全局规范姿态的搜寻定义为一个决策问题,以可学习的方式预测候选姿态集合中最有可能是规范姿态的元素,并将该元素作为输入点云的位姿进行特征提取步骤。具体而言,PRinvNet首先设计姿态扩展器模块,按如下算法构建候选姿态集合:对输入点云应用PCA分析获取3个特征向量,组成全局参考系,初步规范点云姿态。针对全局参考系3个轴方向的模糊性问题,枚举所有可能的方向组合,组成了具有8种姿态可能性的候选姿态集合,而后用有限旋转群进一步扩展候选姿态集合,以提高对不同噪声模式的兼容性。在姿态选择器模块中,网络回归针对候选姿态集合的决策向量,并将向量中最大值对应的集合元素作为输入点云的全局规范姿态。PCA-RINet则深入探讨了“使用PCA算法分析全局规范姿态”这一过程存在的模糊性及其对于特征提取的影响,并设计了一个消歧模块,该模块输入所有候选姿态下的点云坐标情况,用其回归各姿态的权重,以加权平均各个姿态下的点云坐标。
然而,由于点云存在的遮挡、噪点、扰动以及同一类别下不同实例类型等繁杂多样的问题,全局唯一的规范姿态难以计算和选定。上述方法在符号消歧步骤中也存在奇异性和不连续性等问题,并且考虑所有候选姿态会增加网络的计算负担。EIP(efficient invariant pose)(Fei和Deng,2024a)提出一种新型的全局规范姿态学习方法,该网络从输入点云中提取旋转不变的几何属性,并基于这些几何属性学习不变权重,然后使用这些权重计算全局规范姿态。EIP 方法具有可以忽略不计的复杂性和坚实的理论基础,支持简易地集成到任意非旋转不变的点云网络中,为基于全局规范姿态的旋转不变点云网络提供了新的发展思路。
纵观现有主流方法,从旋转不变能力的稳定性来说,基于几何属性的旋转不变点云网络更有优势,局部参考系和全局规范姿态均存在不同噪声模式下估计不一致的问题,而距离、夹角等几何属性相对来说更加鲁棒;从语义信息的感知能力来说,基于局部参考系的旋转不变网络和基于全局规范姿态的旋转不变网络则更为突出,现有的几何属性构建方法,难以像这两者一样保留更为完整的语义信息。不过,随着研究的不断深入,诸如PARI-Conv(pose-aware rotation invariant convolution)、EIP(efficient invariant pose)、EFRI-N(ellipsoid-fitting-based rotation invariant network)(黎书玉 等,2022)、GLC-HCANglobal-local-consistent hypergraph cross-attention network(Dai等,2025)等网络将几何属性、局部参考系和全局规范姿态结合使用,优势互补,以实现性能更强、旋转不变能力更稳定的点云深度学习网络。
旋转等变的点云特征提取网络具有这样一个属性:在输入点云PRN×3P∈RN×3发生旋转R R∈ SO(3)的情况下,输出特征会相应地发生有规律的变换(Deng等,2021a)。即满足
F(RP)=T(R)F(P)
式中,F为旋转等变的点云特征提取网络,T则代表输入点云的旋转和输出特征的变换之间的映射关系,这种映射关系能够由等变网络的设计原理显式确定。因此,旋转等变网络既无需经过信息有损的几何分析预处理步骤,直接输入原始的点云坐标,提取具有形状信息和姿态信息的点云特征,又允许研究者根据等变网络的内部原理,设计变换策略,将姿态信息从特征中剥离,从而获得旋转不变的点云特征。本文将这种旋转不变点云网络的构造方法称为“特征旋转不变方法”,总体逻辑如图8所示。根据前端等变点云网络架构和数学理论的不同,这些方法可细分为5类,分别为:群卷积网络、球面卷积网络、张量场网络、向量网络以及四元数网络。
群卷积神经网络(group equivariant convolutional network,G-CNN)(Cohen和Welling,2016)最初用于解决图像网络中的二维旋转鲁棒性问题,广泛应用于图像分类、目标识别等任务,研究者也将该技术拓展到三维点云网络中。
群卷积的直观理解在于定义一个离散的旋转群,枚举群中每一个旋转姿态并提取该姿态下的点云特征。假设只从该离散群中采样旋转,输入点云的旋转变换则可以被映射为各姿态特征的排列次序,这种基于排列次序堆叠形成的特征矩阵便是群卷积网络得到的旋转等变特征。
之后,研究者对该特征矩阵应用跨姿态的池化、注意力等机制将其转化为旋转不变特征,代表性方法有Rot-SO-Net(rotation self-organizing network)(Li等,2019)、EPN(equivariant point network)(Chen等,2021)和E2PN(efficient equivariant point network)(Zhu等,2023)。其中,Rot-SO-Net首次尝试将群卷积思想引入点云网络中,该网络用共享的特征提取模块提取离散群中各个姿态{Ri}i=0K-1{Ri}i=0K-1下的点云特征FRM×C'F∈RM×C',并进行有序堆叠,得到旋转等变的特征矩阵,如图9所示。之后,Rot-SO-Net沿姿态维度应用最大池化或平均池化等操作,从而将等变特征转化为旋转不变特征FR1×C'F∈R1×C',并应用于下游分类、分割等任务。EPN则首次完善定义了点云网络中的群卷积算子,并针对刚性变换群(special euclidean group in 3 dimensions,SE(3)),设计等变特征提取网络。该网络提出可分离卷积算子,将复杂度较高的SE(3)群卷积分解为两个可分离的卷积运算符,在三维欧氏空间和SO(3)群空间中交替进行特征提取,在不影响性能的情况下显著降低了计算成本。此外,EPN网络还引入了群注意力池化机制,为离散旋转群中的每种姿态计算注意力权重,用该权重聚合各姿态特征,获得更具有表达能力的旋转不变特征。 E2PN进一步优化SE(3)群卷积的效率问题,通过引入商空间(S2×R3S2×R3,其中S2S2表示二维球面空间)减少卷积计算时枚举的离散姿态个数,显著降低点云群卷积计算的存储和计算成本。具体而言,E2PN将SO(3)群离散化后,将平面旋转群(special orthogonal group in 2 dimensions,SO(2))作为稳定子群,对离散SO(3)群求商,进而将原先定义在SO(3)群上的卷积操作降维到商空间上进行,简化了卷积的计算。此外,E2PN设计了一个排列层,以极小的代价从商空间中恢复SO(3)群的等变特征,从而保证网络的旋转等变性。E2PN不仅保持了EPN的高准确率,还显著提升了显存利用率和计算速度。
尽管如此,上述的点云群卷积网络只能保证在离散的旋转群上旋转等变,使得基于池化、注意力等机制获取的旋转不变特征不够稳定,在任意旋转情况的分类、分割任务下,群卷积网络的性能指标会大幅下降。因此,点云群卷积网络通常会随机旋转输入点云进行数据增强训练,缓解连续旋转空间下的低泛化能力缺陷。近期,SE3Conv3D(Weijler和Hermosilla,2025)结合PCA算法分析感受野内部的候选局部参考系集合,并以该集合为离散旋转群进行群卷积局部特征提取,建立了局部等变群卷积点云网络。通过这种与传统几何分析算法相结合的方式,SE3Conv3D的旋转不变性得到了大幅增强。
图像卷积网络中,输入数据和输出特征均是一种平面信号,输入数据发生平移变换,输出特征也会进行相应的平移变换。类似地,球面卷积网络将输入数据转换成一种球形信号(Cohen等,2018)。若输入数据发生了旋转,球面卷积输出的球形特征也会发生相应的旋转变换,使得球面卷积网络符合旋转等变定义。
该种网络最初用于网格(mesh)或体素数据的等变特征提取。其中,S2CNN(spherical convolutional neural network)(Cohen等,2018)用射线投影的方式将网格形状投影到封闭的球体上以构建球面信号输入,并定义球面互相关操作,将球面信号转换为SO(3)信号,提高网络的表达能力,打破了传统球面卷积定义中卷积核关于z轴对称的限制。由于球面相关性满足广义傅里叶定理,S2CNN能够使用广义快速傅里叶变换算法快速提取旋转等变特征。之后,S2CNN网络在最后一层对在SO(3)空间上的特征应用全局最大池化,得到旋转不变特征输出;相反,Spherical-CNN(Esteves等,2018)仅在球面域上实现球面卷积,用球形傅里叶变换将卷积核和球形信号在谱域上展开,并以矩阵乘法的方式在谱域上实现球面卷积。该方法在最后一层引入全局尺度的加权平均池化层,产生旋转不变特征。之后,研究者尝试对点云数据应用球面卷积。PRIN(pointwise rotation-invariant network)(You等,2020)对点云进行球形体素化,将球形体素作为输入,设计球面卷积网络,如图10所示。针对均匀体素化下,点云在体素之间分布不均的问题,PRIN设计密度感知的自适应采样模块进行非均匀的球形体素化,从而保证点云中点在球形体素的均匀分布,便于后续卷积特征提取。之后,PRIN定义球形体素卷积算子,输入球形体素,输出球形体素的逐点旋转不变特征,即体素上每个空间位置的特征只与其形状有关,输入球形体素信号的旋转只会导致输出球形体素特征发生相应的旋转,并不会造成旋转前后对应位置体素的特征向量值发生变化;SPRIN(You等,2022)则是对PRIN的优化工作,实现点云数据上直接进行球形体素卷积,避免了体素化带来的计算开销,同时保留了PRIN中逐点旋转不变的核心机制。
目前,PRIN和SPRIN实现的球形体素卷积逐渐在点云下游任务上得到应用。例如,LEAD(Marcon等,2022)基于球形体素卷积提取局部描述符,用于点云配准;Compass(Spezialetti等,2020)基于球形体素卷积寻找场景点云中局部切片的姿态,即输出局部切片的LRF。但球形体素卷积也通过傅里叶变换进行加速,而傅里叶变换中进行的谱域展开,需要实现基函数离散化,这限制了网络的表达能力。
张量场网络通常使用SO(3)群的不可约表示(irreducible representation,IR)实现旋转等变性。这类方法在欧氏空间中对点云进行特征提取,并在每一层都保证了对旋转、平移和点集排列的等变性。其中,TFN(tensor field network)(Thomas等,2018)首次将不可约表示这一数学理念带入到旋转等变点云网络研究中,其将输入的点云表示为张量场(tensor fields,TF),根据其秩的不同,张量场可以具象化为标量、向量或更高阶的张量。之后,TFN基于球谐函数和可学习的径向基函数设计卷积核,以保持输入输出的等变性。具体而言,TFN的卷积核F(lf,li)(r)F(lf,li)(r)由径向函数R(lf,li)(r)R(lf,li)(r)和球谐函数Y(l)(r^)Y(l)(r^)组成,形式为
F(lf,li)(r)=R(lf,li)(r)Y(l)(r^)
式中,lili和lflf分别表示输入和输出特征的不可约表示阶数,rr表示两点间的相对方向向量,rr则为该方向向量的模,即相对距离,r^r^为单位方向向量。该设计可以捕捉输入点云在不同旋转下多阶特征,提高网络的语义信息感知能力。为了从旋转等变特征中获得旋转不变特征,TFN只使用0阶特征用于分类、分割等任务。这是因为理论上,0阶的不可约表示为实数,对于旋转具有不变性。
SE(3)-Transformer(Fuchs等,2020)是对TFN网络的扩展,它设计不可约表示下的注意力机制以代替TFN的卷积操作。具体而言,不可约表示下的特征内积具有旋转不变性,该网络在点云的邻域上计算邻域点等变特征和关键点等变特征的内积,从而获取旋转不变量作为注意力权重。此外,SE(3)-Transformer对TFN网络中球谐函数的相关计算进行了重构,将计算速度提升了至多3个数量级,增大了张量场网络能够处理的点云数据体量。
SPHNet(spherical harmonics kernels network)(Poulenard等,2019)则是张量场网络的一种变体,该网络同样基于径向基函数和球谐函数设计旋转等变卷积层,不同的是,SPHNet首先结合狄拉克测度(Escobedo,1986)将点云数据扩展到体积信号,再进行旋转等变卷积,最后将结果限制回原始点云以获取各点的旋转等变特征,并将每个特征通道下的二范数定义为旋转不变特征。
不过,张量场网络过于复杂的相关数学理论,限制了该方向的发展。并且,增加不可约表示的阶数会导致网络训练资源消耗成倍增长,使得张量场网络的网络深度和特征维度都比较小,难以扩展。
相比于复杂数学理论支撑的球面卷积网络和张量场网络,向量神经网络仅通过将神经元从标量扩展为三维向量,实现了SO(3)旋转等变性,如图11所示。VNN(Deng等,2021a)首次提出向量神经元的概念,以简化点云旋转与特征变换之间的等变映射关系,并且扩展常见的神经网络操作(如线性层、激活函数和池化层)至向量神经元版本。基于这些神经网络层,VNN实现了具有旋转等变性质的VN-PointNet和VN-DGCNN网络。为将网络应用至分类、分割等任务上,VNN还定义了不变层模块,将向量特征之间的内积作为旋转不变特征,输入到标签预测模块;VN-Transformer(Assaad等,2023)进一步定义向量神经元下的注意力机制,以提高向量网络的特征提取能力。此外,VN-Transformer发现向量线性层在处理范数较小的点时,无法保持数值稳定的分布式训练,于是在向量线性层中添加具有可控范数的偏置参数来提高训练稳定性和模型质量。
近年,一些研究者认为VNN的向量神经元只有三维,这限制了向量网络的特征提取能力。例如,TetraSphere(Melnyk等,2024)将向量神经元的维度提升到四维,并结合三维球形神经元设计TT层(tetra-transform layer),从输入点云中提取旋转等变的四维向量特征矩阵作为初始特征,应用到四维VNN网络中进行后续特征提取;FER-VNN(frequency-based equivariant feature representation for vector neural network)(Son等,2024)则通过传统算法分析和捕捉点云数据中的多个频率信息,从而将三维点云映射到高维频率特征空间,并在该空间上设计更高维度的VNN网络。
向量网络的数学理论难度相对较小,很快得到下游点云任务的青睐,在实例分割(Lei等,2023)、点云配准(Yao等,2025)等任务上均有应用。
四元数网络用四元数表示输入点云和中间特征,相比于传统基于标量的神经网络,四元数网络的特征在面对旋转时自然具有等变属性。在四元数网络gg中,将点云xx和特征ff都用对应的四元数形式xuxu,fvfv表示,则对点云输入应用旋转变换RR相当于对所有中间层的四元数特征应用相同的旋转变换,从而确保在旋转下的特征一致性,如图12所示。
REQNN(rotation-equivariant guaternion neural network)(Shen等,20202024)首次定义了四元数网络的基础结构,并提出一组转换规则,能够将常见的点云网络(如PointNet、DGCNN等)修改为旋转等变的四元数神经网络。之后,REQNN将四元数特征的二范数作为网络输出的旋转不变特征;QMLP-RInv(quaternion multi-layer perceptron for rotation invariant network)(Zhang等,2020b)则设计四元数乘积单元,该单元能够在特征提取过程中解耦出旋转不变特征和旋转等变特征。值得一提的是,QMLP-RInv从理论上证明了四元数乘积单元的合理性和效率;QEC-Net(Zhao等,2020)则结合胶囊网络技术,设计了一个适用于四元数网络的胶囊网络。其中每个胶囊是一个基于动态路由算法(Aftab等,2015)设计的QEC模块(quaternion equivariant capsule module),该模块输入点云的一个特征提取局部邻域和预先计算的若干局部参考系,输出旋转等变的四元数特征和旋转不变的实数特征,以将姿态和形状信息解耦。
四元数网络与向量网络一样具有较为简单的数学理论和实现难度。但从实验性能上,四元数网络在所测试分类任务上性能并不突出,这成为其在下游任务上的研究应用相对向量网络少的一个原因。
特征旋转不变网络是一种后处理方法,借助旋转等变网络能够保留姿态信息的能力,直接在原始点云上提取旋转等变特征,再利用池化、注意力机制以及规范变换等操作剥离等变特征中包含的姿态信息,得到旋转不变特征。相较于几何旋转不变网络这种预处理方法,特征旋转不变网络的主要优势在于接收的信息更为完整,不存在任何类似于几何属性计算的信息抽取步骤,但主要劣势也很明显:1)网络设计较为复杂,需要研究者具有较为深厚的数理基础才能够继续研究,难以与现有其他非旋转不变的优秀点云网络相结合;2)大多数旋转等变网络训练和推理的资源消耗较高,使得特征旋转不变方法较难进一步拓展到更大规模的点云数据上,例如室内场景点云、室外激光雷达扫描点云等;3)大多数等变网络采用池化或计算标量值(例如内积、二范数等)的方式将旋转等变特征转换为旋转不变特征,但是这种方式存在较大的信息丢失,较难应用到语义更为理解细节的任务。
“训练旋转不变”相关方法的核心是利用旋转数据增强与优化目标(训练方案和损失函数)的耦合设计,使非旋转不变网络隐式学习旋转不变特征。以有监督学习为例,其数学形式可定义为:给定输入点云样本PRN×3P∈RN×3,并从三维旋转群SOSO(3)中随机采样旋转矩阵RSOR∈SO(3)对输入点云进行刚性变换,构建增广训练集{Ri×P}i=1K{Ri×P}i=1K,并优化如下目标函数,计算为
minθERSO(3)[L(Fθ(R×P),y)]
式中,LL表示优化目标,一般为各方法的损失函数,yy表示监督标签,FF为非旋转不变网络,θθ表示可训练的网络参数组合。整体优化过程可视做以强制让模型在训练过程中适应随机旋转扰动的方式,驱动网络参数空间向隐式编码旋转不变性方向演化。
即使旋转增强的训练方案具有训练成本高、不稳定的局限性,但其设计简便、易于与前沿非旋转不变点云网络结合的优势无法忽视。因此,一些研究者从非旋转不变点云网络FF、优化目标LL和旋转采样策略3个角度探索效果更好的训练旋转不变方案。
从非旋转不变点云网络FF的模型架构角度出发,早期,PointNet会在特征提取主干之外加入姿态规范器模块T-Net(transformation min-network),将输入点云对齐到规范姿态下进行特征提取,如图13所示。结合旋转增强训练,T-Net模块能有效地为输入点云估计其相应的变换矩阵,并将输入点云转换到规范位姿空间。后续方法也提出各自的T-Net设计,以实现更稳定的规范位姿估计能力,如IT-Net(Yuan等,2019)回归四元数表示,同时规范化旋转和平移;RTN(rotation transformation network)(Deng等,2021b)回归离散的欧拉角,来规范点云位姿等。不过,T-Net模块具有以下局限性:1)点云诸如遮挡、采样精度等噪声严重制约了T-Net模块的有效性,并不能确保输出合适的变换矩阵,使不同姿态下的点云数据对齐至规范姿态;2)即使加入正交矩阵正则化项,也无法保证输出变换矩阵完全正交,这使得输入点云不可避免地发生形变,点云包含的几何信息和语义信息会发生变化,对后续提取旋转不变性特征造成损害。
不同于上述姿态对齐的方式,SPE-Net(selective position encoding network)(Qiu等,2022)选择增加模块以在特征空间弱化姿态相关信息对于输出特征的贡献度。该网络提出一种选择性位置编码策略,结合注意力机制,在特征空间中发现与姿态有关的信息,动态确定网络参数的优化行为,有效地降低训练期间参数优化的自由度,从而通过减少训练难度来更好地利用旋转增强数据,使SPE-Net在训练和测试期间对旋转噪声更具有鲁棒性。
从旋转增强的采样策略出发,ART-Point(adversarial rotation point network)(Wang等,2022)尝试将对抗性训练(adversarial training,AT)技术引入训练方案。该模型基于点云分类这一任务,将点云的旋转视为一种攻击,在训练点云网络的同时,不断提高旋转攻击难度,自适应地产生更难分类的旋转噪声点云。这种网络和攻击难度迭代学习的模式,有效提高了网络的抗旋转能力。
从优化目标LL的设计策略出发,PaRot(Zhang等,2023)引入孪生训练的方式提取旋转不变特征。具体而言,该模型将输入的点云数据划分为若干局部块和全局块,并对每个块施加一对随机旋转操作。这些旋转后的点云块通过共享的网络分支进行编码,而后,PaRot模型将块编码解耦为几何特征向量和方向向量。对于几何特征向量,PaRot设计旋转不变性损失函数,最小化旋转增强点云块间几何特征向量的距离,约束网络学习旋转不变的形状内容;对于方向向量,PaRot设计旋转等变损失函数,约束学习到的方向向量保持与旋转矩阵的等变关系,以确保提取点云块的姿态信息。
总体而言,除了T-Net模块以外,ART-Point、SPE-Net、PaRot模型虽然无法做到理论上完备的旋转不变能力,但从实验结果上,经过各自方案训练后,原本非旋转不变的点云网络能够在任意姿态情况下的分类、分割等任务上保持较为稳定的性能水平。
根据各类旋转不变点云网络相关文献,本节总结主流研究中较为常用的评估任务、数据集、评价指标和测试设置等。总的来看,无论何种任务,旋转不变点云网络的研究大体都设置了如下3种不同的训练/测试模式,对任务内每个评价指标进行3次评估,以更加精确地衡量网络的特征提取能力和抗旋转能力(Zhang等,2019b):
1)Z/Z。表示在仅环绕Z轴采样的旋转数据增强下训练,同样在仅环绕Z轴采样的旋转数据增强下测试。
2)Z/SO(3)。表示在仅环绕Z轴采样的旋转数据增强下训练,但在随机采样SO(3)群的旋转数据增强下测试。
3)SO(3)/SO(3)。表示在随机采样SO(3)群的旋转数据增强下训练,同样在随机采样SO(3)群的旋转数据增强下测试。
点云分类任务是最为常用的旋转不变特征评估任务。该任务下评价指标有总体分类精度(overall accuracy,OA)(Zhang等,2019b)等,指正确分类的点云个数与测试集中点云总数的比值。最常用的数据集为ModelNet40(Wu等,2015)和ScanObjectNN(Uy等,2019)。
ModelNet40数据集由合成的物体数据组成,原始数据为CAD网格(computer-aided design mesh)。该数据集类别多样、形状干净无噪声、数据集构建良好,因此广受欢迎,是使用最为广泛的点云分析基准之一。用于点云网络研究时,研究者通常先从CAD网格表面均匀采样点得到点云数据,再进行如下两步预处理:1)将点云质心对齐至坐标原点;2)将点云缩放至单位球体内。
ScanObjectNN数据集包含对2 902个现实世界物体扫描得来的15 000个带噪声点云,给分类带来更多挑战。研究者基于背景点(background,BG)、随机平移(translation, T)、随机旋转(rotation,R)和随机改变点云尺度(scaling,R)等扰动(perturbation,PB),对物体点云(object,OBJ)施加噪声,划分了5类不同噪声模式的测试子集,分别如下:
1)OBJ_BG表示带有背景噪声点的物体点云;
2)PB_T25表示在25%的限度内对物体点云施加的平移扰动;
3)PB_T25_R表示在25%的限度内对物体点云施加平移扰动,并施加随机旋转;
4)PB_T50_R表示在50%的限度内对物体点云施加平移扰动,并施加随机旋转;
5)PB_T50_RS表示在50%的限度内对物体点云施加平移扰动,并施加随机旋转和随机改变点云大小。
另外一个是普通测试子集OBJ_ONLY(object only)。其中常用的是 OBJ_ONLY、OBJ_BG 和PB_T50_RS 3个测试子集。研究者一般仅展示每个测试子集下的分类精度,并不取整体测试集下的平均分类精度。
点云分割任务可以按点云体量不同细分为:物体点云的分割和场景点云的分割。其中,物体点云的分割又称部件分割(part segmentation),是旋转不变特征评估中最常用的分割任务。该任务下的评价指标有实例级别的交并比(intersection over union at instance-level,IoU(ins.))和类级别的交并比(intersection over union at classification-level,IoU(cls.))等(Lou等,2023),两个指标越大,说明网络在点云分割任务上的性能越好。常用的数据集为ShapeNetPart(Chang等,2015),该数据集包含来自16个类别的16 881个三维物体点云,总共标注了50种部件。
在场景点云上进行点云分割以评估旋转不变点云网络的工作主要面向语义分割任务。该任务下的评价指标有平均交并比(mean intersection over union,mIoU)等(Zhang等,2025)。常用的数据集有S3DIS(the Stanford 3D indoor scene dataset)(Armeni等,2016)等,S3DIS包含6个带有271间客房的大型室内区域扫描点云,点云中的每个点都标注了语义类别,整个数据集总共有13种语义类别。
三维形状检索旨在用输入点云的全局特征,从数据库中检索形状一致的点云。常用评价指标有:检索精度(precision at n,PN)、召回率(recall at n, R@N)、F1分数(F1-score at n, F1@N)、平均精度均值(mean average precision,mAP)和归一化累积增益(rormalized discounted cumulative gain,NDCG)等(Zhang等,2022)。常用的数据集为 ShapeNetCo- re55(Chang等,2015)。该数据集包含55个类别,总共51 190个三维形状,划分为两个子集:正常数据集和扰动数据集。研究者通常会从两个子集的原始网格数据中均匀采样得到点云数据。
依据各类旋转不变点云网络相关文献,本文选用普遍使用的ModelNet40分类数据集、ShapeNetPart分割数据集对各代表性方法进行性能统计和比较,定量统计如表1表2所示。
从旋转不变能力角度来看,除E2PN、SPENet外,其他旋转不变方法的实验性能在Z/Z、Z/SO(3)、SO(3)/SO(3) 3种测试基本稳定,不会出现大幅的性能波动,这是因为E2PN属于离散群卷积方法,其旋转不变能力仅存在于网络所定义的离散旋转群。而SPENet属于训练旋转不变方法,其网络设计只能弱化姿态相关信息对于特征提取的影响,并不能避免影响。从方法性能角度来看,几何旋转不变方法普遍优于特征旋转不变方法和训练旋转不变方法。其中,RISurConv在ModelNet40点云分类数据集上取得了表中所比较方法的最佳水平。SPENet在ShapeNetPart分割数据集上的IoU (ins.)指标上取得了所比较方法的最佳水平。这说明,训练旋转不变方法在感知输入点云的高维语义细节方面更有优势;而特征旋转不变方法因其复杂的特征空间约束,以及不变特征转换过程的信息丢失,总体上性能稍逊于同时期的其他旋转不变方法。
综上所述,目前的几何旋转不变方法更适合类似于分类的识别任务,目前的训练旋转不变方法在感知输入点云的高维语义细节方面表现更好,而目前的特征旋转不变方法则在分类、分割等具有旋转不变性要求的任务上稍逊一筹。依据前文的系统性讨论和定量指标统计,本综述定性归纳总结了几何旋转不变、特征旋转不变和训练旋转不变3类方法的优劣势,如表3所示。
随着旋转不变点云网络的研究逐渐深入,各类下游任务的研究者开始意识到旋转不变特征的价值,并将其应用到自身的研究中,推动着各自领域的发展,包括但不限于以下几个典型的任务:
1)点云自监督表示学习。该任务旨在从未标记的点云数据中学习通用且有意义的点云表示,其中一种主流方法是掩码自重建方法,即对点云进行随机遮蔽,仅从可视部分中学习特征,重建被遮蔽的点云。RI-MAE(rotation-invariant masked autoencoder)(Su等,2024)、MaskLRF(masked autoencoding of local reference frame)(Furuya,2024)在局部参考系下编码可视点云特征,并重建局部参考系下的遮蔽点云,从而实现点云表示的旋转不变性。
2)端到端点云配准。该任务的关键是在不同噪声、不同姿态的待配准点云中,寻找正确的点对匹配关系,因此特征必然需要旋转不变性。最初,端到端点云配准网络利用随机旋转增强来提高模型的旋转鲁棒性,近期,RoITr(rotation invariant Transformer)(Yu等,2023a)、PARE-Net(position-aware rotation equivariant network)(Yao等,2025)分别基于几何属性和向量网络提取旋转不变特征用于点对匹配关系挖掘,两个网络也在任意旋转情况下的配准测试中,取得了优异的性能表现。
3)点云补全。该任务旨在对点云数据本身进行处理,输入的残缺点云,输出补全后的完整点云,该任务需要旋转不变点云特征以确保能够为不同姿态下的同一点云建立形状一致的补全结果。RICNet(rotation invariant completion network)(Chen和Shi,2024)、CF-Net(complementary fusion network)(Chen等,2022)利用基于几何属性的旋转不变点云网络提取特征,再进行后续补全网络的计算和学习,确保补全结果在不同姿态下保持一致。
4)位置重识别。该任务旨在为场景点云切片寻找到其在整个场景中的位置,可视为一种匹配任务。因此与配准任务一样,点云特征也需要具有旋转不变性。RIANet(rotation invariant aware network)(Hao等,2024)、LR-Net(lightweight and rotation invariant network)(Zhang等,2024b)同样基于旋转不变几何属性提取场景点云切片的全局特征,这使得后续的匹配过程不受姿态影响,进一步提高该任务的鲁棒性。
5)三维目标检测。该任务旨在识别输入传感器数据中的物体信息,包括类别、位置和大小等,但场景中物体姿态是各异的,需要旋转不变特征辅助识别场景中目标的类别、大小等位姿无关属性。RIDE(Wang等,2024a)提出双特征提取器,一方面提取姿态敏感但几何信息完善的目标相关特征;另一方面提取旋转不变特征。两种特征相辅相成,在保证基本性能的同时,增强后续目标检测模块对于旋转的鲁棒性。
6)三维目标重建。该任务旨在从输入点云中重建物体原有表面,在数字孪生、虚拟现实领域有广泛应用。现有方法多在已对齐数据上,追求高精度的重建结果,但当输入点云发生旋转时,这些方法的重建精度会显著降低。对此,ART-InvRec(adversarial framework for rotation invariant 3D object reconstruction)(Yang等,2025)与ART-Point(adversarial rotation for point cloud network)方法相似,将点云的旋转视为一种攻击,在训练重建网络的同时,自适应地提高重建网络的旋转鲁棒性。
7)点云生成。该任务旨在基于输入的潜在表征生成高质量的三维点云,在计算机视觉、自动驾驶等领域具有重要应用。SE(3)-Diffusion(Peng等,2024)提出点云潜在表征需要编码点云的形状、纹理等信息,而与姿态信息无关,并将扩散模型与几何先验相结合,通过距离和角度联合建立SE(3)不变的几何属性空间。该方法在该空间中学习点云潜在表征,在保持生成质量的同时提升了点云生成模型对旋转、平移等几何变换的鲁棒性。
本综述回顾了现有旋转不变点云网络的研究历程和重要成果。将相关方法按旋转不变性的来源划分为:几何旋转不变方法、特征旋转不变方法以及训练旋转不变方法,并在各节进一步细分讨论。同时也统计了该领域研究所用的常用下游任务和数据集,调研了旋转不变点云网络的典型下游应用任务,并对未来研究方向进行了展望,希望能为领域研究者提供一些思考:
1)面向实际采样过程的旋转不变性建模。现有旋转不变点云网络研究多使用已有的点云数据库,判断随机旋转点云数据后网络输出特征是否一致,但却忽略了在实际场景中,采集不同姿态物体的点云数据时,由传感器噪声、动态遮挡、非均匀采样等引入的噪声问题。例如,自动驾驶场景中激光雷达受运动畸变影响,同一物体在不同位姿下可能呈现显著的点密度差异与局部缺失,以及不同姿态下物体的自遮挡问题。未来研究可以构建更贴近真实物理采样的旋转增强方法,探索噪声鲁棒性与旋转不变性的联合优化机制以提升模型在复杂真实场景中的泛化能力。
2)不同类型旋转不变策略的相互结合。几何旋转不变、特征旋转不变与训练旋转不变方法在理论完备性、计算复杂度以及语义保留能力上各具优势。未来可以探索如何将不同类型的旋转不变策略相互结合,例如通过对抗性训练辅助局部参考系的自适应纠正,增强噪声鲁棒性,或利用旋转等变特征预测输入点云合适的全局规范姿态等。
3)旋转不变点云网络在多模态大模型中的应用。随着近年来大模型的迅猛发展,多模态大模型也将点云数据囊括,逐步应用于三维视觉领域(Thengane等,2025)。在机器人自主导航、自动驾驶感知等实际场景中,大模型需处理大规模、多视角、动态变化的点云数据,而姿态多样性导致的特征变化问题会显著降低模型在开放环境中的语义感知准确性。现有方法在进行模态对齐时多采用简单的旋转数据增强,以保证一定的鲁棒性,但与旋转不变性还存在一定的距离(Yi等,2023)。因此,未来研究可以探索旋转不变点云网络与多模态大模型的结合,以保证大模型对于物体姿态变化的鲁棒性。
总体来看,该研究领域尚处于探索阶段,除了上述未来展望以及继续提出新的旋转不变点云网络,提高下游任务性能之外,还存在如下几个问题有待进一步研究,以促进该领域深入发展。
1)有效几何属性的区分。局部邻域中可以通过多种方式获得丰富的距离、夹角等信息,目前方法局限于几何属性提取算法的设计,深入讨论深度学习框架下,不同旋转不变几何属性的有效性以及冗余情况能够为该领域的发展提供一些新颖的思路。
2)高可集成性的旋转不变特征提取。目前非旋转不变或旋转等变的点云网络,依然保持着规范姿态数据下的优异性能,如何以简便、不损害原有性能的方式,将旋转不变性集成到这些网络中,是一个值得研究的主题。并且,旋转不变点云网络应具有一定的简洁性和可复用性,方便下游任务应用。
3)高计算效率的不变特征提取。目前许多方法虽然能够取得良好的性能,但是时间开销和计算开销较大,难以处理大规模的点云数据,在下游任务上的应用非常受限。这些方法大多仅能处理小体量的物体点云。在保证特征提取能力的前提下,设计更高效的旋转不变点云网络可能会更快推动点云深度学习的发展。
参考文献 引证文献
排序方式:
Aftab KHartley R and Trumpf J. 2015. Lq-closest-point to affine subspaces using the generalized weiszfeld algorithm. International Journal of Computer Vision114(1): 1-15 [DOI: 10.1007/s11263-014-0791-8
Aoki YGoforth HSrivatsan R A and Lucey S. 2019. PointNetLK: robust and efficient point cloud registration using PointNet//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, USA: IEEE:7156-7165 [DOI: 10.1109/CVPR.2019.00733
Armeni ISener OZamir A RJiang H LBrilakis IFischer M and Savarese S. 2016. 3D semantic parsing of large-scale indoor spaces//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, USA: IEEE:1534-1543 [DOI: 10.1109/CVPR.2016.170
Assaad SDowney CAl-Rfou RNayakanti N and Sapp B. 2023. VN-transformer: rotation-equivariant attention for vector neurons [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2206.04176.pdf
Bello S AAlfasly SMao J WLu JLi LXu C and Zou Y R. 2025. Geometric edge convolution for rigid transformation invariant features in 3D point clouds. Neurocomputing622: #129313 [DOI: 10.1016/j.neucom.2024.129313
Chang A XFunkhouser TGuibas LHanrahan PHuang Q XLi Z MSavarese SSavva MSong S RSu HXiao J XYi L and Yu F. 2015. ShapeNet: an information-rich 3D model repository [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/1512.03012.pdf
Chen B FYeh Y M and Lu Y C. 2022. CF-Net: complementary fusion network for rotation invariant point cloud completion//Proceedings of 2022 ICASSP IEEE International Conference on Acoustics, Speech and Signal Processing. Singapore, Singapore: IEEE:2275-2279 [DOI: 10.1109/ICASSP43922.2022.9746388
Chen CLi G BXu R JChen T SWang M and Lin L. 2019. ClusterNet: deep hierarchical cluster network with rigorously rotation-invariant representation for point cloud analysis//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, USA: IEEE:4989-4997 [DOI: 10.1109/CVPR.2019.00513
Chen H WLiu S CChen W KLi H and Hill R. 2021. Equivariant point network for 3D point cloud analysis//Proceedings of 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Nashville, USA: IEEE:14509-14518 [DOI: 10.1109/CVPR46437.2021.01428
Chen R H and Cong Y. 2022. The devil is in the pose: ambiguity-free 3D rotation-invariant learning via pose-aware convolution//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, USA: IEEE:7462-7471 [DOI: 10.1109/CVPR52688.2022.00732
Chen Y and Shi P C. 2024. Rotation-invariant completion network//Proceedings of the 6th Chinese Conference on Pattern Recognition and Computer Vision. Xiamen, China: Springer:115-127 [DOI: 10.1007/978-981-99-8432-9_10
Chen Y YDuan L HZhao S SDing C X and Tao D C. 2024. Local-consistent transformation learning for rotation-invariant point cloud analysis//Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE:5418-5427 [DOI: 10.1109/CVPR52733.2024.00518
Choy CPark J and Koltun V. 2019. Fully convolutional geometric features//Proceedings of 2019 IEEE/CVF International Conference on Computer Vision. Seoul, Korea (South): IEEE:8957-8965 [DOI: 10.1109/ICCV.2019.00905
Cohen T and Welling M. 2016. Group equivariant convolutional networks//Proceedings of the 33rd International Conference on Machine Learning. New York, USA: PMLR:2990-2999
Cohen T SGeiger MKöhler J and Welling M. 2018. Spherical CNNs [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/1801.10130.pdf
Dai YYing S H and Gao Y. 2025. Exploring local and global consistent correlation on hypergraph for rotation invariant point cloud analysis. IEEE Transactions on Multimedia27: 186-197 [DOI: 10.1109/TMM.2024.3521678
Deng C YLitany ODuan Y QPoulenard ATagliasacchi A and Guibas L J. 2021a. Vector neurons: a general framework for SO(3)- equivariant networks//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:12180-12189 [DOI: 10.1109/ICCV48922.2021.01198
Deng H WBirdal T and Ilic S. 2018a. PPF-FoldNet: unsupervised learning of rotation invariant 3D local descriptors//Proceedings of the 15th European Conference on Computer Vision. Munich Germany: Springer:620-638 [DOI: 10.1007/978-3-030-01228-1_37
Deng H WBirdal T and Ilic S. 2018b. PPFNet: global context aware local features for robust 3D point matching//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA: IEEE:195-205 [DOI: 10.1109/CVPR.2018.00028
Deng SLiu BDong Q L and Hu Z Y. 2021b. Rotation transformation network: learning view-invariant point cloud for classification and segmentation//Proceedings of 2021 IEEE International Conference on Multimedia and Expo. Shenzhen, China: IEEE:1-6 [DOI: 10.1109/ICME51207.2021.9428265
Drost BUlrich MNavab N and Ilic S. 2010. Model globally, match locally: efficient and robust 3D object recognition//Proceedings of 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Francisco, USA: IEEE:998-1005 [DOI: 10.1109/CVPR.2010.5540108
Escobedo M. 1986. On a characterisation of Dirac measures. Proceedings of the Royal Society of Edinburgh: Section A Mathematics103(3/4): 253-264 [DOI: 10.1017/S0308210500018904
Esteves CAllen-Blanchette CMakadia A and Daniilidis K. 2018. Learning SO(3) equivariant representations with spherical CNNs//Proceedings of the 15th European Conference on Computer Vision. Munich, Germany: Springer:54-70 [DOI: 10.1007/978-3-030-01261-8_4
Fei J J and Deng Z D. 2024a. Incorporating rotation invariance with non-invariant networks for point clouds//Proceedings of 2024 International Conference on 3D Vision. Davos, Switzerland: IEEE:985-994 [DOI: 10.1109/3DV62453.2024.00070
Fei J J and Deng Z D. 2024b. Rotation invariance and equivariance in 3D deep learning: a survey. Artificial Intelligence Review57(7): #168 [DOI: 10.1007/s10462-024-10741-2
Fuchs F BWorrall D EFischer V and Welling M. 2020. SE(3)-transformers: 3D roto-translation equivariant attention networks//Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc.:1970-1981
Furuya T. 2024. MaskLRF: self-supervised pretraining via masked autoencoding of local reference frames for rotation-invariant 3D point set analysis. IEEE Access12: 73340-73353 [DOI: 10.1109/ACCESS.2024.3404016
Gu R BWu Q XLi Y QKang W XNg W W Y and Wang Z Y. 2022. Enhanced local and global learning for rotation-invariant point cloud representation. IEEE MultiMedia29(4): 24-37 [DOI: 10.1109/MMUL.2022.3151906
Hao WZhang W J and Su H N. 2024. RIA-Net: rotation invariant aware 3D point cloud for large-scale place recognition. IEEE Robotics and Automation Letters9(6): 5014-5021 [DOI: 10.1109/LRA.2024.3384887
He CZhao Z JZhang X BYu H and Wang R H. 2025. RotInv-PCT: rotation-invariant point cloud transformer via feature separation and aggregation. Neural Networks185: #107223 [DOI: 10.1016/j.neunet.2025.107223
Huang Z TYu Y KXu J WNi F and Le X Y. 2020. PF-Net: point fractal network for 3D point cloud completion//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE:7659-7667 [DOI: 10.1109/CVPR42600.2020.00768
Kim SPark J and Han B. 2020. Rotation-invariant local-to-global representation learning for 3D point cloud//Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc.:8174-8185
Kwon SHur J H and Kim H. 2023. Deep learning-based point cloud upsampling: a review of recent trends. JMST Advances5(4): 105-111 [DOI: 10.1007/s42791-023-00058-6
Lei J HDeng C YSchmeckpeper KGuibas L and Daniilidis K. 2023. EFEM: equivariant neural field expectation maximization for 3D object segmentation without scene supervision//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver, Canada: IEEE:4902-4912 [DOI: 10.1109/CVPR52729.2023.00475
Li F RFujiwara KOkura F and Matsushita Y. 2021. A closer look at rotation-invariant deep point cloud analysis//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:16198-16207 [DOI: 10.1109/ICCV48922.2021.01591
Li G BZhang R FChen C and Lin L. 2022. Rotation-invariant deep hierarchical cluster network for point cloud analysis. Journal of Software33(11): 4356-4378
李冠彬, 张锐斐, 陈超, 林倞. 2022a. 基于旋转不变深度层次聚类网络的点云分析. 软件学报33(11): 4356-4378 [DOI: 10.13328/j.cnki.jos.006315
Li J XBi Y C and Lee G H. 2019. Discrete rotation equivariance for point cloud recognition//Proceedings of 2019 International Conference on Robotics and Automation. Montreal, Canada: IEEE:7269-7275 [DOI: 10.1109/ICRA.2019.8793983
Li S YZhang X DSun R and Fan Z G. 2022. Point cloud rotation invariant network based on ellipsoid fitting. Journal of Electronic Measurement and Instrumentation36(9): 111-117
黎书玉, 张旭东, 孙锐, 范之国. 2022b. 基于椭球拟合的点云旋转不变网络. 电子测量与仪器学报36(9): 111-117 [DOI: 10.13382/j.jemi.B2205143
Li X ZLi R HChen G YFu C WCohen-Or D and Heng P A. 2022. A rotation-invariant framework for deep point cloud analysis. IEEE Transactions on Visualization and Computer Graphics28(12): 4503-4514 [DOI: 10.1109/TVCG.2021.3092570
Li Z ZLi GLi T HLiu S and Gao W. 2023. Semantic point cloud upsampling. IEEE Transactions on Multimedia25: 3432-3442 [DOI: 10.1109/TMM.2022.3160604
Liang DGuo Y CZhang S KZhang S HHall PZhang M and Hu S. 2018. LineNet: a zoomable CNN for crowdsourced high definition maps modeling in urban environments [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/1807.05696.pdf
Lin J JRickert M and Knoll A. 2021. Deep hierarchical rotation invariance learning with exact geometry feature representation for point cloud classification//Proceedings of 2021 IEEE International Conference on Robotics and Automation. Xi’an, China: IEEE:9529-9535 [DOI: 10.1109/ICRA48506.2021.9561307
Lou Y JYe Z LYou YJiang N JLu J BWang W MMa L Z and Lu C W. 2023. CRIN: rotation-invariant point cloud analysis and rotation estimation via centrifugal reference frame//Proceedings of the 37th AAAI Conference on Artificial Intelligence. Washington, USA: AAAI:1817-1825 [DOI: 10.1609/aaai.v37i2.25271
Luo S TLi J HGuan J QSu Y FCheng C RPeng J and Ma J Z. 2022. Equivariant point cloud analysis via learning orientations for message passing//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, USA: IEEE:18910-18919 [DOI: 10.1109/CVPR52688.2022.01836
Ma XQin CYou H XRan H X and Fu Y. 2022. Rethinking network design and local geometry in point cloud: a simple residual MLP framework [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2202.07123.pdf
Mahler JLiang JNiyaz SLaskey MDoan RLiu X YOjea J A and Goldberg K. 2017. Dex-Net 2.0: deep learning to plan robust grasps with synthetic point clouds and analytic grasp metrics [EB/OL]. [2017-08-08]. https://arxiv.org/pdf/1703.09312.pdf
Marcon MSpezialetti RSalti SSilva L and Di Stefano L. 2022. Unsupervised learning of local equivariant descriptors for point clouds. IEEE Transactions on Pattern Analysis and Machine Intelligence44(12): 9687-9702 [DOI: 10.1109/TPAMI.2021.3126713
Melnyk PRobinson AFelsberg M and Wadenbäck M. 2024. TetraSphere: a neural descriptor for O(3)-invariant point cloud analysis//Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE:5620-5630 [DOI: 10.1109/CVPR52733.2024.00537
Peng YTang L LLiao QLiu YQi S H and Zhang J J. 2024. SE(3)- Diffusion: an equivariant diffusion model for 3D point cloud generation//Proceedings of the 15th International Conference on Genetic and Evolutionary Computing. Kaohsiung, China: Springer:510-522 [DOI: 10.1007/978-981-97-0068-4_48
Poulenard ARakotosaona M JPonty Y and Ovsjanikov M. 2019. Effective rotation-invariant point CNN with spherical harmonics kernels//Proceedings of 2019 International Conference on 3D Vision. Quebec City, Canada: IEEE:47-56 [DOI: 10.1109/3DV.2019.00015
Qi C RSu HMo K and Guibas L J. 2017a. PointNet: deep learning on point sets for 3D classification and segmentation//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, USA: IEEE:77-85 [DOI: 10.1109/CVPR.2017.16
Qi C RYi LSu H and Guibas L J. 2017b. PointNet++: deep hierarchical feature learning on point sets in a metric space//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, USA: Curran Associates Inc.:5105-5114
Qin ZYu HWang C JGuo Y LPeng Y XIlic SHu D W and Xu K. 2023. GeoTransformer: fast and robust point cloud registration with geometric transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence45(8): 9806-9821 [DOI: 10.1109/TPAMI.2023.3259038
Qiu Z FLi Y HWang YPan Y WYao T and Mei T. 2022. SPE-Net: boosting point cloud analysis via rotation robustness enhancement//Proceedings of the 17th European Conference on Computer Vision. Tel Aviv, Israel: Springer:593-609 [DOI: 10.1007/978-3-031-20062-5_34
Rusu R BBlodow N and Beetz M. 2009. Fast point feature histograms (FPFH) for 3D registration//Proceedings of 2009 IEEE International Conference on Robotics and Automation. Kobe, Japan: IEEE:3212-3217 [DOI: 10.1109/ROBOT.2009.5152473
Shen WWei Z HRen Q HZhang B BHuang S KFan J Q and Zhang Q S. 2024. Interpretable rotation-equivariant quaternion neural networks for 3D point cloud processing. IEEE Transactions on Pattern Analysis and Machine Intelligence46(5): 3290-3304 [DOI: 10.1109/TPAMI.2023.3346383
Shen WZhang B BHuang S KWei Z H and Zhang Q S. 2020. 3D-rotation-equivariant quaternion neural networks//Proceedings of the 16th European Conference on Computer Vision. Glasgow, UK: Springer:531-547 [DOI: 10.1007/978-3-030-58565-5_32
Son DKim JSon S and Kim B. 2024. An intuitive multi-frequency feature representation for SO(3)-equivariant networks [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2405.04537.pdf
Spezialetti RStella FMarcon MSilva LSalti S and Di Stefano L. 2020. Learning to orient surfaces by self-supervised spherical CNNs//Proceedings of the 34th International Conference on Neural Information Processing Systems. Vancouver, Canada: Curran Associates Inc.:5381-5392
Su K MWu Q XCai P PZhu X GLu X QWang Z Y and Hu K. 2024. RI-MAE: rotation-invariant masked AutoEncoders for self-supervised point cloud representation learning [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2409.00353.pdf
Thengane VZhu X TBouzerdoum SPhung S L and Li Y P. 2025. Foundational models for 3D point clouds: a survey and outlook [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2501.18594.pdf
Thomas H. 2020. Rotation-invariant point convolution with multiple equivariant alignments//Proceedings of 2020 International Conference on 3D Vision. Fukuoka, Japan: IEEE:504-513 [DOI: 10.1109/3DV50981.2020.00060
Thomas NSmidt TKearnes SYang L SLi LKohlhoff K and Riley P. 2018. Tensor field networks: rotation- and translation-equivariant neural networks for 3D point clouds [EB/OL]. [2018-05-18]. https://arxiv.org/pdf/1802.08219.pdf
Tombari FSalti S and Di Stefano L. 2010. Unique signatures of histograms for local surface description//Proceedings of the 11th European Conference on Computer Vision. Heraklion, Greece: Springer:356-369 [DOI: 10.1007/978-3-642-15558-1_26
Uy M APham Q HHua B SNguyen T and Yeung S K. 2019. Revisiting point cloud classification: a new benchmark dataset and classification model on real-world data//Proceedings of 2019 IEEE/CVF International Conference on Computer Vision. Seoul, Korea (South): IEEE:1588-1597 [DOI: 10.1109/ICCV.2019.00167
Vaswani AShazeer NParmar NUszkoreit JJones LGomez A NKaiser Ł and Polosukhin I. 2017. Attention is all you need//Proceedings of the 31st International Conference on Neural Information Processing Systems. Long Beach, USA: Curran Associates Inc.:6000-6010
Wang R BYang Y B and Tao D C. 2022. ART-point: improving rotation robustness of point cloud classifiers via adversarial rotation//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans, USA: IEEE:14351-14360 [DOI: 10.1109/CVPR52688.2022.01397
Wang YSun Y BLiu Z WSarma S EBronstein M M and Solomon J M. 2019. Dynamic graph CNN for learning on point clouds. ACM Transactions on Graphics38(5): #146 [DOI: 10.1145/3326362
Wang Z XHan XLiu H X and Li X Z. 2024a. RIDE: boosting 3D object detection for LiDAR point clouds via rotation-invariant analysis [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2408.15643.pdf
Wang Z XYu Y L and Li X Z. 2024b. Rethinking local-to-global representation learning for rotation-invariant point cloud analysis. Pattern Recognition154: #110624 [DOI: 10.1016/j.patcog.2024.110624
Weijler L and Hermosilla P. 2025. Efficient continuous group convolutions for local SE(3) equivariance in 3D point clouds [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/2502.07505.pdf
Wu Z RSong S RKhosla AYu FZhang L GTang X O and Xiao J X. 2015. 3D ShapeNets: a deep representation for volumetric shapes//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition. Boston, USA: IEEE:1912-1920 [DOI: 10.1109/CVPR.2015.7298801
Xu J YTang XZhu Y SSun J and Pu S L. 2021. SGMNet: learning rotation-invariant point cloud representations via sorted Gram matrix//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:10448-10457 [DOI: 10.1109/ICCV48922.2021.01030
Yan YMao Y X and Li B. 2018. SECOND: sparsely embedded convolutional detection. Sensors18(10): #3337 [DOI: 10.3390/s18103337
Yang X YYang RHu FFang LZhong WChen Z Y and Wang G. 2025. ART-InvRec: an adversarial framework for rotation-invariant 3D object reconstruction. Displays87: #102915 [DOI: 10.1016/j.displa.2024.102915
Yao R ZDu S YCui W TTang C H and Yang C W. 2025. PARE-Net: position-aware rotation-equivariant networks for robust point cloud registration//Proceedings of the 18th European Conference on Computer Vision. Milan, Italy: Springer:287-303 [DOI: 10.1007/978-3-031-72904-1_17
Yi X YDeng J JSun Q RHua X SLim J H and Zhang H W. 2023. Invariant training 2D-3D joint hard samples for few-shot point cloud recognition//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision. Paris, France: IEEE:14417-14428 [DOI: 10.1109/ICCV51070.2023.01330
You YLou Y JLiu QTai Y WMa L ZLu C W and Wang W M. 2020. Pointwise rotation-invariant network with adaptive sampling and 3D spherical voxel convolution//Proceedings of the 34th AAAI Conference on Artificial Intelligence. New York, USA: AAAI:12717-12724 [DOI: 10.1609/aaai.v34i07.6965
You YLou Y JShi R XLiu QTai Y WMa L ZWang W M and Lu C W. 2022. PRIN/SPRIN: on extracting point-wise rotation invariant features. IEEE Transactions on Pattern Analysis and Machine Intelligence44(12): 9489-9502 [DOI: 10.1109/TPAMI.2021.3130590
Yu HQin ZHou JSaleh MLi D SBusam B and Ilic S. 2023a. Rotation-invariant transformer for point cloud matching//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver, Canada: IEEE:5384-5393 [DOI: 10.1109/CVPR52729.2023.00521
Yu JLin YWang BYe Q and Cai J Q. 2019. An advanced outlier detected total least-squares algorithm for 3-D point clouds registration. IEEE Transactions on Geoscience and Remote Sensing57(7): 4789-4798 [DOI: 10.1109/TGRS.2019.2893054
Yu J HZhang C Y and Cai W D. 2023b. Rethinking rotation invariance with point cloud registration//Proceedings of the 37th AAAI Conference on Artificial Intelligence. Washington, USA: AAAI:3313-3321 [DOI: 10.1609/aaai.v37i3.25438
Yu R XWei XTombari F and Sun J. 2020. Deep positional and relational feature learning for rotation-invariant point cloud analysis//Proceedings of the 16th European Conference on Computer Vision. Glasgow, UK: Springer:217-233 [DOI: 10.1007/978-3-030-58607-2_13
Yu X MRao Y MWang Z YLiu Z YLu J W and Zhou J. 2021. PoinTr: diverse point cloud completion with geometry-aware transformers//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:12478-12487 [DOI: 10.1109/ICCV48922.2021.01227
Yuan W THeld DMertz C and Hebert M. 2019. Iterative transformer network for 3D point cloud [EB/OL]. [2025-01-23]. https://arxiv.org/pdf/1811.11209.pdf
Zhang D XYu J HZhang C Y and Cai W D. 2023. PaRot: patch-wise rotation-invariant network via feature disentanglement and pose restoration//Proceedings of the 37th AAAI Conference on Artificial Intelligence. Washington, USA: AAAI:3418-3426 [DOI: 10.1609/aaai.v37i3.25450
Zhang HWang C SYu LTian S WNing X and Rodrigues J. 2024a. PointGT: a method for point-cloud classification and segmentation based on local geometric transformation. IEEE Transactions on Multimedia26: 8052-8062 [DOI: 10.1109/TMM.2024.3374580
Zhang J MYu M YVasudevan R and Johnson-Roberson M. 2020a. Learning rotation-invariant representations of point clouds using aligned edge convolutional neural networks//Proceedings of 2020 International Conference on 3D Vision. Fukuoka, Japan: IEEE:200-209 [DOI: 10.1109/3DV50981.2020.00030
Zhang STong H HXu J J and Maciejewski R. 2019a. Graph convolutional networks: a comprehensive review. Computational Social Networks6(1): #11 [DOI: 10.1186/s40649-019-0069-y
Zhang XQin S FXu Y and Xu H T. 2020b. Quaternion product units for deep learning on 3D rotation groups//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, USA: IEEE:7302-7311 [DOI: 10.1109/CVPR42600.2020.00733
Zhang Z HLiu HWang XShu M CChen G L and Zhang Q Z. 2024b. Lightweight and rotation-invariant place recognition network for large-scale raw point clouds. ISPRS Journal of Photogrammetry and Remote Sensing212: 58-72 [DOI: 10.1016/j.isprsjprs.2024.04.030
Zhang X LFu C L and Zhang Y J. 2020. Extended pointwise convolution network model for point cloud classification and segmentation. Journal of Image and Graphics25(8): 1551-1557
张新良, 付陈琳, 赵运基. 2020. 扩展点态卷积网络的点云分类分割模型. 中国图象图形学报25(8): 1551-1557 [DOI: 10.11834/jig.190508
Zhang Z YHua B SChen WTian Y B and Yeung S K. 2020c. Global context aware convolutions for 3D point cloud understanding//Proceedings of 2020 International Conference on 3D Vision. Fukuoka, Japan: IEEE:210-219 [DOI: 10.1109/3DV50981.2020.00031
Zhang Z YHua B SRosen D W and Yeung S K. 2019b. Rotation invariant convolutions for 3D point clouds deep learning//Proceedings of 2019 International Conference on 3D Vision. Quebec City, Canada: IEEE:204-213 [DOI: 10.1109/3DV.2019.00031
Zhang Z YHua B S and Yeung S K. 2022. RIConv++: effective rotation invariant convolutions for 3D point clouds deep learning. International Journal of Computer Vision130(5): 1228-1243 [DOI: 10.1007/s11263-022-01601-z
Zhang Z YYang L C and Xiang Z Y. 2025. RISurConv: rotation invariant surface attention-augmented convolutions for 3D point cloud classification and segmentation//Proceedings of the 18th European Conference on Computer Vision. Milan, Italy: Springer:93-109 [DOI: 10.1007/978-3-031-73390-1_6
Zhao CYang J QXiong XZhu A FCao Z G and Li X. 2022. Rotation invariant point cloud analysis: where local geometry meets global topology. Pattern Recognition127: #108626 [DOI: 10.1016/j.patcog.2022.108626
Zhao H SJiang LJia J YTorr P and Koltun V. 2021. Point transformer//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision. Montreal, Canada: IEEE:16239-16248 [DOI: 10.1109/ICCV48922.2021.01595
Zhao Y HBirdal TLenssen J EMenegatti EGuibas L and Tombari F. 2020. Quaternion equivariant capsule networks for 3D point clouds//Proceedings of the 16th European conference on computer vision. Glasgow, UK: Springer:1-19 [DOI: 10.1007/978-3-030-58452-8_1
Zhu M HGhaffari MClark W A and Peng H E. 2023. E2PN: efficient SE(3)-equivariant point network//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Vancouver, Canada: IEEE:1223-1232 [DOI: 10.1109/CVPR52729.2023.00124
2025年第30卷第12期
PDF下载
149
71
引用本文
BibTeX
文章信息
doi: 10.11834/jig.250029
  • 接收时间:2025-01-23
  • 首发时间:2026-04-09
  • 出版时间:2025-12-16
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2025-01-23
  • 修回日期:2025-04-25
基金
作者信息
    1西北工业大学计算机学院,西安710072

通讯作者:

参考文献
分享链接
https://castjournals.cast.org.cn/joweb/zgtxtxxb/CN/10.11834/jig.250029
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

Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)

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