PointJEM: Self-supervised Point Cloud Understanding for Reducing Feature Redundancy via Joint Entropy Maximization

Simple item page
dc.contributor.authorCao, Xinen_US
dc.contributor.authorXia, Huanen_US
dc.contributor.authorWang, Haoyuen_US
dc.contributor.authorSu, Linzhien_US
dc.contributor.authorZhou, Pingen_US
dc.contributor.authorLi, Kangen_US
dc.contributor.editorChen, Renjieen_US
dc.contributor.editorRitschel, Tobiasen_US
dc.contributor.editorWhiting, Emilyen_US
dc.date.accessioned2024-10-13T18:03:11Z
dc.date.available2024-10-13T18:03:11Z
dc.date.issued2024
dc.description.abstractMost deep learning methods for point cloud processing are supervised and require extensive labeled data. However, labeling point cloud data is a tedious and time-consuming task. Self-supervised representation learning can solve this problem by extracting robust and generalized features from unlabeled data. Yet, the features from representation learning are often redundant. Current methods typically reduce redundancy by imposing linear correlation constraints. In this paper, we introduce PointJEM, a self-supervised representation learning method for point clouds. It includes an embedding scheme that divides the vector into parts, each learning a unique feature. To minimize redundancy, PointJEM maximizes joint entropy between parts, making the features pairwise independent. We tested PointJEM on various datasets and found it significantly reduces redundancy beyond linear correlation. Additionally, PointJEM performs well in downstream tasks like classification and segmentation.en_US
dc.description.sectionheadersPoint Cloud Processing and Analysis I
dc.description.seriesinformationPacific Graphics Conference Papers and Posters
dc.identifier.doi10.2312/pg.20241276
dc.identifier.isbn978-3-03868-250-9
dc.identifier.pages10 pages
dc.identifier.urihttps://doi.org/10.2312/pg.20241276
dc.identifier.urihttps://diglib.eg.org/handle/10.2312/pg20241276
dc.publisherThe Eurographics Associationen_US
dc.rightsAttribution 4.0 International License
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/
dc.subjectKeywords: point cloud, representation learning, self-supervised; CCS Concepts: Computing methodologies → Computer vision representations; Networks → Network design principles
dc.subjectpoint cloud
dc.subjectrepresentation learning
dc.subjectself supervised
dc.subjectComputing methodologies → Computer vision representations
dc.subjectNetworks → Network design principles
dc.titlePointJEM: Self-supervised Point Cloud Understanding for Reducing Feature Redundancy via Joint Entropy Maximizationen_US
Files
Original bundle
Now showing 1 - 1 of 1
Thumbnail Image
Name:
pg20241276.pdf
Size:
7.59 MB
Format:
Adobe Portable Document Format
Download
Collections
PG2024 Conference Papers and Posters