StraightPCF: Straight Point Cloud Filtering

TL;DR

StraightPCF reframes point-cloud denoising as transporting noisy patches toward the clean surface along straight paths, learning a constant flow velocity $\pmb{v}_\theta$ and a distance scalar $d_\phi$ to scale the flow. The method employs a coupled VelocityModule architecture (two modules, $K=2$) plus a DistanceModule to straighten trajectories and ensure convergence, resulting in a lightweight network of about $5.3\times 10^5$ parameters. It achieves state-of-the-art Chamfer Distance and Point-to-Mesh results on synthetic (PUNet/PCNet) and real-world data (Kinect, Rue-Madame) without requiring regularization or post-processing, and it produces well-distributed filtered points. The approach offers efficient inference and strong generalization, making it practical for downstream 3D vision tasks without heavy computational demands.

Abstract

Point cloud filtering is a fundamental 3D vision task, which aims to remove noise while recovering the underlying clean surfaces. State-of-the-art methods remove noise by moving noisy points along stochastic trajectories to the clean surfaces. These methods often require regularization within the training objective and/or during post-processing, to ensure fidelity. In this paper, we introduce StraightPCF, a new deep learning based method for point cloud filtering. It works by moving noisy points along straight paths, thus reducing discretization errors while ensuring faster convergence to the clean surfaces. We model noisy patches as intermediate states between high noise patch variants and their clean counterparts, and design the VelocityModule to infer a constant flow velocity from the former to the latter. This constant flow leads to straight filtering trajectories. In addition, we introduce a DistanceModule that scales the straight trajectory using an estimated distance scalar to attain convergence near the clean surface. Our network is lightweight and only has $\sim530K$ parameters, being 17% of IterativePFN (a most recent point cloud filtering network). Extensive experiments on both synthetic and real-world data show our method achieves state-of-the-art results. Our method also demonstrates nice distributions of filtered points without the need for regularization. The implementation code can be found at: https://github.com/ddsediri/StraightPCF.

Figures (14)

• Figure 1: Filtered trajectories for the Isocahedron shape at 50K resolution and noise scale $\sigma=3\%$. Our StraightPCF filters points along much straighter paths, compared to ScoreDenoise Luo-Score-Based-Denoising.
• Figure 2: Our StraightPCF network. It involves a coupled VelocityModule stack that infers a constant flow velocity $\pmb{v}^k_\theta$ for patch states $\Tilde{\pmb{X}}_{(\hat{t}+k)/T}$. To ensure filtered points converge to the surface, the DistanceModule infers a distance scalar $d_\phi$ that scales the velocity.
• Figure 3: Our StraightPCF is able to recover a better distribution of filtered points, even at very high noise scales ($\sigma=3\%$ and 50K resolution) unseen during network training.
• Figure 4: StraightPCF models initial noisy patches (light blue) as being intermediate states of a linear interpolation between high noise variants (red) and the clean surfaces (dark blue) and encourages straight filtering trajectories.
• Figure 5: Left: Filtering by coupled VelocityModules only. Right: Coupled VelocityModules and DistanceModule. Scaled trajectories (green lines) lead to better convergence at the surface.