PointMixer: MLP-Mixer for Point Cloud Understanding

MLP-Mixer has newly appeared as a new challenger against the realm of CNNs and Transformer. Despite its simplicity compared to Transformer, the concept of channel-mixing MLPs and token-mixing MLPs achieves noticeable performance in image recognition tasks. Unlike images, point clouds are inherently...

Full description

Saved in:
Bibliographic Details
Published inComputer Vision - ECCV 2022 Vol. 13687; pp. 620 - 640
Main Authors Choe, Jaesung, Park, Chunghyun, Rameau, Francois, Park, Jaesik, Kweon, In So
Format Book Chapter
LanguageEnglish
Published Switzerland Springer 01.01.2022
Springer Nature Switzerland
SeriesLecture Notes in Computer Science
Online AccessGet full text
ISBN9783031198113
3031198115
ISSN0302-9743
1611-3349
DOI10.1007/978-3-031-19812-0_36

Cover

Abstract MLP-Mixer has newly appeared as a new challenger against the realm of CNNs and Transformer. Despite its simplicity compared to Transformer, the concept of channel-mixing MLPs and token-mixing MLPs achieves noticeable performance in image recognition tasks. Unlike images, point clouds are inherently sparse, unordered and irregular, which limits the direct use of MLP-Mixer for point cloud understanding. To overcome these limitations, we propose PointMixer, a universal point set operator that facilitates information sharing among unstructured 3D point cloud. By simply replacing token-mixing MLPs with Softmax function, PointMixer can “mix” features within/between point sets. By doing so, PointMixer can be broadly used for intra-set, inter-set, and hierarchical-set mixing. We demonstrate that various channel-wise feature aggregation in numerous point sets is better than self-attention layers or dense token-wise interaction in a view of parameter efficiency and accuracy. Extensive experiments show the competitive or superior performance of PointMixer in semantic segmentation, classification, and reconstruction against Transformer-based methods.
AbstractList MLP-Mixer has newly appeared as a new challenger against the realm of CNNs and Transformer. Despite its simplicity compared to Transformer, the concept of channel-mixing MLPs and token-mixing MLPs achieves noticeable performance in image recognition tasks. Unlike images, point clouds are inherently sparse, unordered and irregular, which limits the direct use of MLP-Mixer for point cloud understanding. To overcome these limitations, we propose PointMixer, a universal point set operator that facilitates information sharing among unstructured 3D point cloud. By simply replacing token-mixing MLPs with Softmax function, PointMixer can “mix” features within/between point sets. By doing so, PointMixer can be broadly used for intra-set, inter-set, and hierarchical-set mixing. We demonstrate that various channel-wise feature aggregation in numerous point sets is better than self-attention layers or dense token-wise interaction in a view of parameter efficiency and accuracy. Extensive experiments show the competitive or superior performance of PointMixer in semantic segmentation, classification, and reconstruction against Transformer-based methods.
Author Choe, Jaesung
Park, Chunghyun
Rameau, Francois
Kweon, In So
Park, Jaesik
Author_xml – sequence: 1
  givenname: Jaesung
  orcidid: 0000-0002-8978-6702
  surname: Choe
  fullname: Choe, Jaesung
  email: jaesung.choe@kaist.ac.kr
– sequence: 2
  givenname: Chunghyun
  orcidid: 0000-0001-9743-8495
  surname: Park
  fullname: Park, Chunghyun
– sequence: 3
  givenname: Francois
  orcidid: 0000-0001-5031-7653
  surname: Rameau
  fullname: Rameau, Francois
– sequence: 4
  givenname: Jaesik
  orcidid: 0000-0001-5541-409X
  surname: Park
  fullname: Park, Jaesik
– sequence: 5
  givenname: In So
  orcidid: 0000-0001-9626-5983
  surname: Kweon
  fullname: Kweon, In So
BookMark eNpVUMtOAzEMDFAQbekfcNgfCNjxbpxwQ1V5SK3ogZ6jfaTQUu2W7Fbi80m3XDjZmvHYnhmJQd3UXohbhDsE4HvLRpIEQonWoJLgSJ-JSYQpgj0G52KIGlESpfbiH4c0EEMgUNJySldihJSBSjljuhaTtt0CgOI4q2AoaNls6m6x-fHhIVnMl7Jvk3UTkp5JprvmUCWruvKh7fK62tQfN-Jyne9aP_mrY7F6mr1PX-T87fl1-jiXWwLbSW0g9zbl-IpVbLTlqip0agjXHlOrSsqMtd5UaONvORnWaIqssJpLH63RWKjT3nYf4lkfXNE0X61DcMeYXPTsyEXXro_EHWOKovQk2ofm--DbzvmjqvR1F_Jd-Znvu-jEMSpmQJdpcJo1_QIWzmQ1
ContentType Book Chapter
Copyright The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
Copyright_xml – notice: The Author(s), under exclusive license to Springer Nature Switzerland AG 2022
DBID FFUUA
DEWEY 006.37
DOI 10.1007/978-3-031-19812-0_36
DatabaseName ProQuest Ebook Central - Book Chapters - Demo use only
DatabaseTitleList
DeliveryMethod fulltext_linktorsrc
Discipline Applied Sciences
Computer Science
EISBN 9783031198120
3031198123
EISSN 1611-3349
Editor Farinella, Giovanni Maria
Avidan, Shai
Cissé, Moustapha
Brostow, Gabriel
Hassner, Tal
Editor_xml – sequence: 1
  fullname: Avidan, Shai
– sequence: 2
  fullname: Cissé, Moustapha
– sequence: 3
  fullname: Farinella, Giovanni Maria
– sequence: 4
  fullname: Brostow, Gabriel
– sequence: 5
  fullname: Hassner, Tal
EndPage 640
ExternalDocumentID EBC7127701_560_676
GroupedDBID 38.
AABBV
AAZWU
ABSVR
ABTHU
ABVND
ACBPT
ACHZO
ACPMC
ADNVS
AEDXK
AEJLV
AEKFX
AHVRR
ALMA_UNASSIGNED_HOLDINGS
BBABE
CZZ
FFUUA
IEZ
SBO
TPJZQ
TSXQS
Z5O
Z7R
Z7S
Z7U
Z7W
Z7X
Z7Y
Z7Z
Z81
Z82
Z83
Z84
Z85
Z87
Z88
-DT
-~X
29L
2HA
2HV
ACGFS
ADCXD
EJD
F5P
LAS
LDH
P2P
RSU
~02
ID FETCH-LOGICAL-j309t-680ae9470309278697ddb64831fe1492c35899e8d19757a387618b5b967ce6113
ISBN 9783031198113
3031198115
ISSN 0302-9743
IngestDate Tue Jul 29 20:26:52 EDT 2025
Thu May 29 17:02:14 EDT 2025
IsPeerReviewed true
IsScholarly true
LCCallNum TA1634
Language English
LinkModel OpenURL
MergedId FETCHMERGED-LOGICAL-j309t-680ae9470309278697ddb64831fe1492c35899e8d19757a387618b5b967ce6113
Notes Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/978-3-031-19812-0_36.
J. Choe and C. Park—Both authors have equally contributed to this work.
OCLC 1350247573
ORCID 0000-0001-5541-409X
0000-0001-5031-7653
0000-0001-9626-5983
0000-0002-8978-6702
0000-0001-9743-8495
PQID EBC7127701_560_676
PageCount 21
ParticipantIDs springer_books_10_1007_978_3_031_19812_0_36
proquest_ebookcentralchapters_7127701_560_676
PublicationCentury 2000
PublicationDate 2022-01-01
PublicationDateYYYYMMDD 2022-01-01
PublicationDate_xml – month: 01
  year: 2022
  text: 2022-01-01
  day: 01
PublicationDecade 2020
PublicationPlace Switzerland
PublicationPlace_xml – name: Switzerland
– name: Cham
PublicationSeriesTitle Lecture Notes in Computer Science
PublicationSeriesTitleAlternate Lect.Notes Computer
PublicationSubtitle 17th European Conference, Tel Aviv, Israel, October 23-27, 2022, Proceedings, Part XXVII
PublicationTitle Computer Vision - ECCV 2022
PublicationYear 2022
Publisher Springer
Springer Nature Switzerland
Publisher_xml – name: Springer
– name: Springer Nature Switzerland
RelatedPersons Hartmanis, Juris
Gao, Wen
Steffen, Bernhard
Bertino, Elisa
Goos, Gerhard
Yung, Moti
RelatedPersons_xml – sequence: 1
  givenname: Gerhard
  surname: Goos
  fullname: Goos, Gerhard
– sequence: 2
  givenname: Juris
  surname: Hartmanis
  fullname: Hartmanis, Juris
– sequence: 3
  givenname: Elisa
  surname: Bertino
  fullname: Bertino, Elisa
– sequence: 4
  givenname: Wen
  surname: Gao
  fullname: Gao, Wen
– sequence: 5
  givenname: Bernhard
  orcidid: 0000-0001-9619-1558
  surname: Steffen
  fullname: Steffen, Bernhard
– sequence: 6
  givenname: Moti
  orcidid: 0000-0003-0848-0873
  surname: Yung
  fullname: Yung, Moti
SSID ssj0002731120
ssj0002792
Score 2.524876
Snippet MLP-Mixer has newly appeared as a new challenger against the realm of CNNs and Transformer. Despite its simplicity compared to Transformer, the concept of...
SourceID springer
proquest
SourceType Publisher
StartPage 620
Title PointMixer: MLP-Mixer for Point Cloud Understanding
URI http://ebookcentral.proquest.com/lib/SITE_ID/reader.action?docID=7127701&ppg=676
http://link.springer.com/10.1007/978-3-031-19812-0_36
Volume 13687
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
link http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwnV1Li9swEBbZ9FJ66Ju-8aF7Mi62bEtWoYfFpCwhWRa6G_YmLEdms21jqG3a7rV_pL-lv6yjl-2ke9leTDJRJHlGaB6ab4TQ21SUUSkSEZRVAQ5KRnHAkpQF6iSXSsFkqUMXyxNyfJ7ML9KLyeTXKGupa8W78vpGXMn_SBVoIFeFkr2FZPtOgQCfQb7wBAnDc8_43Q2zmroC9j4Gf6Xh4b5LW4j9WZ6vfBziAYNxaRL-5oVsOquqLOLZHLkD8fJnN4DCiq-y6JxdW9abZv8vqqfN5_F6O60323a5-SH1vZPLxWmgv6g0xsMcHx6FuoGff6k7hU8cQWrMxqYKLjcfFvZI46RudaaY37-l3YTGUQqM96IULkqpUrBVL5--b9prA2fecWhBoUYRyyKDT3XALti0we0xJGn2aaKqL8am2qndewkOR2qcmCpQ_2iIcVIIDBao0WC2PCYH6IBmyRTdOZrNF6s-UAf2HdikYa_eVcVFczRlZqUAQ27WqSnpNLzFCKx505A7bs3eSbw2cM4eoHsK9OIpNAow-yGayO0jdN-6KJ7lfQMkJw9He4ySQfLvvV7uHsj9z2_9k6dl7u3I_Ak6_zg7y48DexdHcBWHrA1IFhaSJUo_MEwzwuh6LUiSxVElwcnGZZyC5y6zdcRoSosYlGyUiVQwQksJwoqfoum23spnyFNV4cCKlhUB1wMsJ0GJqCR0xiQrwoo8R4FjCtcZAzZNuTQsaDiNMKVhxMFY54RCe99xjqvmDXeluIHlPObAcq5ZzhXLX9yq9Ut0d1jNr9C0_dbJ12CFtuKNXSd_AdEWd5E
linkProvider Library Specific Holdings
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.title=Computer+Vision+%E2%80%93+ECCV+2022&rft.au=Choe%2C+Jaesung&rft.au=Park%2C+Chunghyun&rft.au=Rameau%2C+Francois&rft.au=Park%2C+Jaesik&rft.atitle=PointMixer%3A+MLP-Mixer+for%C2%A0Point+Cloud+Understanding&rft.series=Lecture+Notes+in+Computer+Science&rft.date=2022-01-01&rft.pub=Springer+Nature+Switzerland&rft.isbn=9783031198113&rft.issn=0302-9743&rft.eissn=1611-3349&rft.spage=620&rft.epage=640&rft_id=info:doi/10.1007%2F978-3-031-19812-0_36
thumbnail_s http://utb.summon.serialssolutions.com/2.0.0/image/custom?url=https%3A%2F%2Febookcentral.proquest.com%2Fcovers%2F7127701-l.jpg