Abstract

Discovering inter-point connection for efficient high-dimensional feature extraction from point coordinate is a key challenge in processing point cloud. Most existing methods focus on designing efficient local feature extractors while ignoring global connection, or vice versa. In this paper, we design a new Inductive Bias-aided Transformer (IBT) method to learn 3D inter-point relations, which considers both local and global attentions. Specifically, considering local spatial coherence, local feature learning is performed through Relative Position Encoding and Attentive Feature Pooling. We incorporate the learned locality into the Transformer module. The local feature affects value component in Transformer to modulate the relationship between channels of each point, which can enhance self-attention mechanism with locality based channel interaction. We demonstrate its superiority experimentally on classification and segmentation tasks. The code is available at: https://github.com/jiamang/IBT

Original languageEnglish
Title of host publicationProceedings, 2023 IEEE International Conference on Multimedia and Expo Workshops
Subtitle of host publicationICMEW 2023
PublisherIEEE
Pages140-145
Number of pages6
ISBN (Electronic)9798350313154
ISBN (Print)9798350313161 (Print on demand)
DOIs
Publication statusPublished - 2023
Event2023 IEEE International Conference on Multimedia and Expo Workshops, ICMEW 2023 - Brisbane Convention and Exhibition Centre, Brisbane, Australia
Duration: 10 Jul 202314 Jul 2023
https://docs.google.com/document/d/e/2PACX-1vT8M3lRSSv4zjkdkies-joQXODFlz57YylfQDRaIgsdPSh-EkIs9m09ge7JvKy29w/pub (Program)
https://ieeexplore.ieee.org/xpl/conhome/10221939/proceeding (Proceedings)

Publication series

NameProceedings - 2023 IEEE International Conference on Multimedia and Expo Workshops, ICMEW 2023

Conference

Conference2023 IEEE International Conference on Multimedia and Expo Workshops, ICMEW 2023
Country/TerritoryAustralia
CityBrisbane
Period10/07/2314/07/23
OtherThe IEEE International Conference on Multimedia & Expo (ICME) has been the flagship multimedia conference for the past two decades. Through the IEEE societies, the Conference serves as a forum to promote the latest advances in multimedia technologies, systems, and applications from both a research and development perspective.

ICME attracts well over 1000 submissions and 500 participants each year, serving as the prime forum for the dissemination of knowledge in the multimedia field. ICME 2023 will showcase high quality oral and poster presentations, as well as feature Workshops sponsored by IEEE societies. Researchers, developers and practitioners are welcomed to organise such Workshops on any new or emerging topic of Multimedia technology. An exposition of multimedia products, animations and industries will be held in conjunction with the conference. Moreover, proposals for Panels, Tutorials, Special Sessions, Industry Technology Workshops and Grand Challenges are also invited. In ICME 2023, exceptional papers and contributors will be also selected and recognised with prestigious awards.
Internet address

Fingerprint

Dive into the research topics of 'Exploiting inductive bias in transformer for point cloud classification and segmentation'. Together they form a unique fingerprint.

Cite this