Reproduce “GS-ROR$^2$: Bidirectional-guided 3DGS and SDF for Reflective Object Relighting and 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
Zuo-Liang Zhu, Beibei Wang, Jian Yang
GS-ROR2 integrates 3D Gaussian Splatting (3DGS) and Signed Distance Fields (SDF) with bidirectional guidance to enable efficient relighting and high-quality geometry reconstruction of reflective objects from multi-view images. The framework achieves photorealistic results and robust meshes, training in approximately 1.5 hours and rendering at over 200 FPS, demonstrating superior performance over existing GS-based inverse rendering methods.
Key insights
- Bidirectional, geometry-only mutual supervision between 3DGS and SDF effectively combines rendering efficiency and robust geometry without the computational burden of full volumetric rendering.
- Deferred Gaussian splatting, when coupled with SDF-guided geometric constraints, enables realistic and sharp specular reflections on challenging reflective surfaces.
- SDF-aware pruning, which dynamically adjusts based on SDF values, is effective in mitigating 'floater' artifacts in 3DGS representations, improving geometric accuracy.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.