P2P-Bridge: Diffusion Bridges for 3D Point Cloud Denoising

TL;DR

This paper introduces P2P-Bridge, a novel point cloud denoising framework using diffusion bridges that learns optimal transport plans between point clouds, outperforming existing methods on multiple datasets.

Contribution

It adapts diffusion Schrödinger bridges for point cloud denoising, a new approach that learns transport plans rather than predicting point displacements, improving denoising performance.

Findings

Significant improvements over existing methods on PU-Net, ScanNet++, and ARKitScenes.

Incorporating additional features like color or DINOv2 enhances denoising results.

The method achieves strong results using only point coordinates.

Abstract

In this work, we tackle the task of point cloud denoising through a novel framework that adapts Diffusion Schr\"odinger bridges to points clouds. Unlike previous approaches that predict point-wise displacements from point features or learned noise distributions, our method learns an optimal transport plan between paired point clouds. Experiments on object datasets like PU-Net and real-world datasets such as ScanNet++ and ARKitScenes show that P2P-Bridge achieves significant improvements over existing methods. While our approach demonstrates strong results using only point coordinates, we also show that incorporating additional features, such as color information or point-wise DINOv2 features, further enhances the performance. Code and pretrained models are available at https://p2p-bridge.github.io.