Reproduce “Normal-GS: 3D Gaussian Splatting with Normal-Involved Rendering” 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

Meng Wei, Qianyi Wu, Jianmin Zheng, Hamid Rezatofighi, Jianfei Cai

Normal-GS connects surface normals to rendered color in Gaussian splatting by predicting diffuse shading from normals and locally shared illumination, and modeling highlights through normal-dependent reflection directions. This design lets image supervision directly affect normals. On NeRF Synthetic, it reports normal mean absolute error of 20.71, versus 23.56 for GaussianShader; this benchmark does not establish equivalent gains on real-world scenes.

Key insights

  • Explicitly involving surface normals in the 3DGS rendering equation, through the Integrated Directional Illumination Vector (IDIV), allows gradient signals to simultaneously optimize both appearance and geometric accuracy.
  • Modeling local incident lighting with anchor-based regularization for IDIVs provides a more robust and efficient solution for complex scenes compared to relying on global environment maps.
  • A self-regularizing depth-normal loss term effectively encourages Gaussians to form coherent surfaces and align with true surface normals during optimization.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.