Adapting Point Transformer V3 for Point Cloud Denoising
School Name
Dutch Fork High School
Grade Level
10th Grade
Presentation Topic
Computer Science
Presentation Type
Mentored
Abstract
A point cloud is a collection of millions of points scanned on an object's surface, and denoising these points is an essential process for the reconstruction of accurate 3D models, with important applications in autonomous driving, robotic navigation, and medical imaging. Point Transformer V3 (PTv3) is a transformer-based model designed for point cloud processing, but it was not originally developed for denoising tasks. In this project, we adapt PTv3 specifically for point cloud denoising to improve its ability to detect and remove noise from 3D data. We further improve performance by optimizing the implementation for modern versions of PyTorch, CUDA, and GPU hardware. These modifications increase both computational efficiency and denoising accuracy, enabling faster and more precise large-scale 3D data processing.
Recommended Citation
Wang, Michael, "Adapting Point Transformer V3 for Point Cloud Denoising" (2026). South Carolina Junior Academy of Science. 25.
https://scholarexchange.furman.edu/scjas/2026/all/25
Location
Furman Hall 204
Start Date
3-28-2026 9:45 AM
Presentation Format
Oral Only
Group Project
No
Adapting Point Transformer V3 for Point Cloud Denoising
Furman Hall 204
A point cloud is a collection of millions of points scanned on an object's surface, and denoising these points is an essential process for the reconstruction of accurate 3D models, with important applications in autonomous driving, robotic navigation, and medical imaging. Point Transformer V3 (PTv3) is a transformer-based model designed for point cloud processing, but it was not originally developed for denoising tasks. In this project, we adapt PTv3 specifically for point cloud denoising to improve its ability to detect and remove noise from 3D data. We further improve performance by optimizing the implementation for modern versions of PyTorch, CUDA, and GPU hardware. These modifications increase both computational efficiency and denoising accuracy, enabling faster and more precise large-scale 3D data processing.
