High-Fidelity Surface Splatting-Based 3D Reconstruction from Multi-View Images

TL;DR

This paper presents a novel surface splatting-based 3D reconstruction method that improves high-frequency detail preservation and geometric fidelity from multi-view images using a new polynomial kernel and stochastic regularization.

Contribution

It introduces a compact polynomial kernel for IMLS that better captures high-frequency details and incorporates stochastic regularization with Laplacian filtering for enhanced surface reconstruction.

Findings

Achieves state-of-the-art results in surface reconstruction accuracy.

Produces sharper visuals and more detailed geometry from multi-view images.

Outperforms existing methods like 3DGS and NeRF in experiments.

Abstract

Multi-view mesh reconstruction remains a core challenge in computer graphics and vision, especially for recovering high-frequency geometry from sparse observations. Recent methods such as 3D Gaussian Splatting (3DGS) and Neural Radiance Fields (NeRF) rely on post-processing for mesh extraction, thereby limiting joint optimization of geometry and appearance. Implicit Moving Least Squares (IMLS) instead enables direct conversion of point clouds into signed distance and texture fields, supporting end-to-end reconstruction and rendering. However, existing IMLS formulations use exponential kernels that struggle with high-frequency detail. We introduce a compact polynomial kernel with local support and greater flexibility, allowing better control over frequency content and improved geometric fidelity. To further enhance fine details, we incorporate stochastic regularization with Laplacian filtering. Together, these improve the preservation of high-frequency structure while maintaining stable optimization. Experiments show state-of-the-art performance in both surface reconstruction and rendering, yielding more accurate geometry and sharper visuals from multi-view data.