Diffusion-Driven Inter-Outer Surface Separation for Point Clouds with Open Boundaries

TL;DR

This work tackles the TSDF-induced double-surface artifact by proposing a diffusion-driven, post-hoc inter/outer shell separation method for open-boundary point clouds. The approach simulates a diffusion ball inside a hollow point cloud, using an escape boundary to separate inter-layer collisions from outer-surface interactions, and generates new spawn points to explore the geometry, finishing with a Poisson surface reconstruction of the extracted inter-layer points. It delivers a robust single-layer mesh for both watertight and open-boundary geometries, demonstrated on scenes with up to 20{,}000 inter and 20{,}000 outer points and runtime on the order of tens of seconds, with parallel CPU acceleration. While effective, the method acknowledges limitations in outer false detections in perforated inter-layers and incomplete inter-layer capture near sharp corners, positioning it as a lightweight, post-hoc module rather than a replacement for full variational or learning-based pipelines.

Abstract

We propose a diffusion-based algorithm for separating the inter and outer layer surfaces from double-layered point clouds, particularly those exhibiting the "double surface artifact" caused by truncation in Truncated Signed Distance Function (TSDF) fusion during indoor or medical 3D reconstruction. This artifact arises from asymmetric truncation thresholds, leading to erroneous inter and outer shells in the fused volume, which our method addresses by extracting the true inter layer to mitigate challenges like overlapping surfaces and disordered normals. We focus on point clouds with \emph{open boundaries} (i.e., sampled surfaces with topological openings/holes through which particles may escape), rather than point clouds with \emph{missing surface regions} where no samples exist. Our approach enables robust processing of both watertight and open-boundary models, achieving extraction of the inter layer from 20,000 inter and 20,000 outer points in approximately 10 seconds. This solution is particularly effective for applications requiring accurate surface representations, such as indoor scene modeling and medical imaging, where double-layered point clouds are prevalent, and it accommodates both closed (watertight) and open-boundary surface geometries. Our goal is \emph{post-hoc} inter/outer shell separation as a lightweight module after TSDF fusion; we do not aim to replace full variational or learning-based reconstruction pipelines.

Figures (5)

• Figure 1: Diffusion-based inter-layer point cloud separation and reconstruction. (a) TSDF-fused stomach point cloud with a double-surface artifact (inter/outer shells) and topological openings. (b) Particle diffusion simulation: the initial spawn point (red sphere), newly generated spawn points (blue spheres), collided interior points (blue points), and the escape boundary enclosing the cloud (dashed sphere). (c) Extracted inter-layer point cloud. (d) Example surface mesh reconstructed from the extracted inter-layer points (see Sec. \ref{['sec:surface_reconstruction']}).
• Figure 2: Diffusion algorithm visualization. The point cloud represents a hollow object with inter surface points (red) and outer surface points (black). Purple balls marked 0, 1, 2, and 3 represent the initial spawn point, reflected point, free-moving point, and escape point, respectively. The gray dashed sphere is the escape boundary sphere centered at the initial spawn point (0).
• Figure 3: Performance metrics of the diffusion-based algorithm of closed double layer ball 3D model(20000 inter layer points and 20000 outer layer points) with $R_{ball}=2 R_0$. (a) Duplication rate $R_{dup}(i)$ versus step $i$. (b) The $R_{outer}$(red line) and $R_{inter}$(green line) versus step $i$. The $R_{inter}$ curve is fitted by the exponential saturation model $R_{inter}(i)=A_0\,(1-\exp(-i/\tau))$ with $A_0=0.970$ and $\tau=6424.3$.
• Figure 4: Performance metrics of the diffusion-based algorithm of opened double layer ball 3D model(20000 inter layer points and 20000 outer layer points) with $R_{ball}=3 R_0$. (a) $N_{escape}$ versus step $i$. (b) The $R_{outer}$(red line) and $R_{inter}$(green line) versus step $i$. The $R_{inter}$ and $R_{outer}$ are fitted by exponentially saturation function.
• Figure 5: Comparison of surface reconstruction methods for the stomach point cloud. Our method: Preserves hole features, generates a clean single-layer mesh, and maintains smooth input geometries. Poisson method: Produces inconsistent normals, double-layered surfaces, erratic merges, imaginary shapes, and irregular textures. Ball Pivoting method: Leads to undesirable hole bridging.