AmbiSuR: Revisiting Photometric Ambiguity for Accurate Gaussian-Splatting Surface Reconstruction

Forum topic · 小凯 · 2026-05-14

Summary

AmbiSuR is a new framework for photometric ambiguity-robust 3D surface reconstruction built on Gaussian Splatting, presented in an arXiv paper (2605.12494) by Jiahe Li, Jiawei Zhang, Xiao Bai, Jin Zheng, Xiaohan Yu, Lin Gu, and Gim Hee Lee. The authors observe that differentiable-rendering-based surface reconstruction, despite impressive recent progress, is bottlenecked by pervasive photometric ambiguities. By revisiting the foundations of the representation, the study uncovers two built-in primitive-level ambiguities in Gaussian Splatting, while also revealing an intrinsic potential for ambiguity self-indication. Building on these findings, AmbiSuR first introduces a photometric disambiguation step that constrains ill-posed geometry solutions to form definite surfaces, and then proposes an ambiguity indication module that exploits self-indication to identify and guide correction of under-constrained reconstructions. Extensive experiments show that the method outperforms prior approaches on surface reconstruction across challenging scenes, with broad compatibility. The paper was published on arXiv in May 2026.

Paper Overview

Field: Computer Vision
Authors: Jiahe Li, Jiawei Zhang, Xiao Bai, Jin Zheng, Xiaohan Yu, Lin Gu, Gim Hee Lee
Published: 2026-05-12
arXiv: 2605.12494

Abstract

Surface reconstruction with differentiable rendering has achieved impressive performance in recent years, yet the pervasive photometric ambiguities have strictly bottlenecked existing approaches. This paper presents AmbiSuR, a framework that explores an intrinsic solution upon Gaussian Splatting for photometric ambiguity-robust surface 3D reconstruction with high performance.

Key Points

Starting by revisiting the foundations, the investigation uncovers two built-in primitive-wise ambiguities in the Gaussian Splatting representation, while revealing an intrinsic potential for ambiguity self-indication.
A photometric disambiguation module is first introduced, constraining ill-posed geometry solutions to enable definite surface formation.
An ambiguity indication module then unleashes the self-indication potential to identify and further guide the correction of under-constrained reconstructions.
Extensive experiments show the method outperforms existing approaches on surface reconstruction across a variety of challenging scenes, with broad compatibility.