Geometrically aware transformer for point cloud analysis
Abstract With the increasing use of 3D point cloud data in autonomous driving, robotic perception, and remote sensing, efficient and accurate point cloud analysis remains a critical challenge. This study presents PointGA, a lightweight Transformer-based model that enhances geometric perception for i...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-05-01
|
| Series: | Scientific Reports |
| Online Access: | https://doi.org/10.1038/s41598-025-00789-7 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850273031569539072 |
|---|---|
| author | Siyuan Chen Zhiwei Fang Siyao Wan Ting Zhou Chunlin Chen Meng Wang Qianming Li |
| author_facet | Siyuan Chen Zhiwei Fang Siyao Wan Ting Zhou Chunlin Chen Meng Wang Qianming Li |
| author_sort | Siyuan Chen |
| collection | DOAJ |
| description | Abstract With the increasing use of 3D point cloud data in autonomous driving, robotic perception, and remote sensing, efficient and accurate point cloud analysis remains a critical challenge. This study presents PointGA, a lightweight Transformer-based model that enhances geometric perception for improved feature extraction and representation. First, PointGA expands the original 3D coordinates into various geometric information, introducing more prior knowledge into the network. Second, a trigonometric position encoding suitable for point clouds is designed, which effectively enhances the expressive capability of positional information and performs preliminary feature extraction through pooling layers, significantly improving the model’s robustness across various tasks. Finally, a positional differential self-attention (PDA) mechanism with linear complexity is developed to optimize feature representation and achieve efficient computation. Experimental results demonstrate that PointGA achieves 87.6% overall accuracy on the ScanObjectNN dataset for classification and 66.2% mean intersection over union(mIoU) on the S3DIS Area 5 dataset for segmentation, outperforming existing methods. These findings highlight the model’s capability to balance efficiency and accuracy, offering a promising solution for point cloud analysis tasks. |
| format | Article |
| id | doaj-art-888afb7dee524562800895ea2116e31e |
| institution | OA Journals |
| issn | 2045-2322 |
| language | English |
| publishDate | 2025-05-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Scientific Reports |
| spelling | doaj-art-888afb7dee524562800895ea2116e31e2025-08-20T01:51:38ZengNature PortfolioScientific Reports2045-23222025-05-0115111710.1038/s41598-025-00789-7Geometrically aware transformer for point cloud analysisSiyuan Chen0Zhiwei Fang1Siyao Wan2Ting Zhou3Chunlin Chen4Meng Wang5Qianming Li6School of Information Science and Engineering, Hunan Institute of Science and TechnologySchool of Information Science and Engineering, Hunan Institute of Science and TechnologySchool of Information Science and Engineering, Hunan Institute of Science and TechnologySchool of Information Science and Engineering, Hunan Institute of Science and TechnologyHunan Institute of Science and Technology, College of Mechanical EngineeringDepartment of Engineering Surveying, Yueyang Institute of Water Resources and Hydropower Survey Planning and Design Co., LTD.School of Information Science and Engineering, Hunan Institute of Science and TechnologyAbstract With the increasing use of 3D point cloud data in autonomous driving, robotic perception, and remote sensing, efficient and accurate point cloud analysis remains a critical challenge. This study presents PointGA, a lightweight Transformer-based model that enhances geometric perception for improved feature extraction and representation. First, PointGA expands the original 3D coordinates into various geometric information, introducing more prior knowledge into the network. Second, a trigonometric position encoding suitable for point clouds is designed, which effectively enhances the expressive capability of positional information and performs preliminary feature extraction through pooling layers, significantly improving the model’s robustness across various tasks. Finally, a positional differential self-attention (PDA) mechanism with linear complexity is developed to optimize feature representation and achieve efficient computation. Experimental results demonstrate that PointGA achieves 87.6% overall accuracy on the ScanObjectNN dataset for classification and 66.2% mean intersection over union(mIoU) on the S3DIS Area 5 dataset for segmentation, outperforming existing methods. These findings highlight the model’s capability to balance efficiency and accuracy, offering a promising solution for point cloud analysis tasks.https://doi.org/10.1038/s41598-025-00789-7 |
| spellingShingle | Siyuan Chen Zhiwei Fang Siyao Wan Ting Zhou Chunlin Chen Meng Wang Qianming Li Geometrically aware transformer for point cloud analysis Scientific Reports |
| title | Geometrically aware transformer for point cloud analysis |
| title_full | Geometrically aware transformer for point cloud analysis |
| title_fullStr | Geometrically aware transformer for point cloud analysis |
| title_full_unstemmed | Geometrically aware transformer for point cloud analysis |
| title_short | Geometrically aware transformer for point cloud analysis |
| title_sort | geometrically aware transformer for point cloud analysis |
| url | https://doi.org/10.1038/s41598-025-00789-7 |
| work_keys_str_mv | AT siyuanchen geometricallyawaretransformerforpointcloudanalysis AT zhiweifang geometricallyawaretransformerforpointcloudanalysis AT siyaowan geometricallyawaretransformerforpointcloudanalysis AT tingzhou geometricallyawaretransformerforpointcloudanalysis AT chunlinchen geometricallyawaretransformerforpointcloudanalysis AT mengwang geometricallyawaretransformerforpointcloudanalysis AT qianmingli geometricallyawaretransformerforpointcloudanalysis |