Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: Jun 8, 2025

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
04:48

Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

Published on: July 5, 2024

369

Learning Cross-Attention Point Transformer With Global Porous Sampling.

Yueqi Duan, Haowen Sun, Juncheng Yan

    IEEE Transactions on Image Processing : a Publication of the IEEE Signal Processing Society
    |November 1, 2024
    PubMed
    Summary

    This study introduces CrossPoints, a novel point-based transformer using parametric Global Porous Sampling (GPS) for more diverse point cloud tokens. This approach enhances relational information capture, improving performance in 3D tasks.

    Related Concept Videos

    You might also read

    Related Articles

    Articles linked to this work by shared authors, journal, and citation graph.

    Sort by
    Same author

    Idiopathic mesenteric phlebosclerosis: A case report.

    Medicine·2026
    Same author

    Fabrication of water-in-oil high internal phase emulsions containing astringent compounds for enhanced oral sensation.

    Food & function·2026
    Same author

    Amino acid metabolism modulates macrophage polarization: implications for autoimmune-related diseases.

    Frontiers in immunology·2026
    Same author

    ReconX: Reconstruct Any Scene From Sparse Views With Video Diffusion Model.

    IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026
    Same author

    Gut microbiota drives the metabolic dysregulation in obesity-prone individuals by impairing GDCA-mediated activation of brown adipose thermogenesis and ileal GLP-1 secretion.

    Acta pharmaceutica Sinica. B·2026
    Same author

    How periodic decorations in extra-long highway tunnels affect driving safety: A simulation study.

    Traffic injury prevention·2026

    Area of Science:

    • Computer Vision
    • Machine Learning
    • 3D Data Processing

    Background:

    • Transformers are vital for capturing token correlations in point cloud processing.
    • Existing methods like Farthest Point Sampling (FPS) limit token diversity by generating holistic sub-clouds.
    • This limitation hinders the full exploitation of point cloud flexibility for attention mechanisms.

    Purpose of the Study:

    • To introduce CrossPoints, a point-based cross-attention transformer.
    • To propose parametric Global Porous Sampling (GPS) and Complementary GPS (C-GPS) strategies for generating diverse tokens.
    • To enhance the relational information capture in point cloud transformers.

    Main Methods:

    • Developed a cross-attention module utilizing parametric GPS and C-GPS strategies.

    More Related Videos

    A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
    04:23

    A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

    Published on: April 21, 2023

    1.7K
    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
    03:31

    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

    Published on: December 15, 2023

    473

    Related Experiment Videos

    Last Updated: Jun 8, 2025

    Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
    04:48

    Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique

    Published on: July 5, 2024

    369
    A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
    04:23

    A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

    Published on: April 21, 2023

    1.7K
    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
    03:31

    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

    Published on: December 15, 2023

    473
  • Employed evenly-sampled points as queries and GPS/C-GPS sampled points as keys and values.
  • Incorporated a deformable operation for adaptive point adjustment to improve token diversity.

    • Main Results:

    • Demonstrated that FPS is a degenerate case of GPS.
    • Showcased improved learning of structure and geometry relational information through consecutive cross-attention.
    • Achieved significant performance gains on shape classification and indoor scene segmentation tasks.

    Conclusions:

    • The proposed CrossPoints with GPS strategy significantly boosts performance on point cloud tasks.
    • GPS and C-GPS strategies effectively generate diversified tokens, enhancing transformer capabilities.
    • The deformable operation further improves token diversity and model adaptability.