Reproduce “Revisiting Photometric Ambiguity for Accurate Gaussian-Splatting Surface Reconstruction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiahe Li, Jiawei Zhang, Xiao Bai, Jin Zheng, Xiaohan Yu, Lin Gu, Gim Hee Lee
AmbiSuR reconstructs 3D surfaces from multi-view photographs by truncating weakly constrained Gaussian tails, discouraging inconsistent colors along camera rays, and using view-dependent Spherical Harmonics to select primitives for local depth-normal guidance. On DTU, its full design scores 0.461 Cf-Dist, versus 0.477 for uniform guidance. This supports selective regularization in this setup, not a universal uncertainty rule.
Key insights
- Photometric ambiguities within Gaussian Splatting are not only external but also arise intrinsically from the primitive representation and its photometric supervision, impacting geometry accuracy.
- The inherent characteristics of Gaussian primitives, specifically their low-opacity tails and the pixel-wise blending loss, contribute to distinct 'primitive edge' and 'photometric blending' ambiguities that hinder definitive surface formation.
- Spherical Harmonics coefficients, typically used for view-dependent appearance, can also function as an effective, self-indicating signal for photometric ambiguity, allowing for precise identification and targeted regularization of problematic regions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.