Reproduce “Robust Gaussian Splatting” 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

François Darmon, Lorenzo Porzi, Samuel Rota-Bulò, Peter Kontschieder

Robust Gaussian Splatting (RGS) models capture imperfections in 3D Gaussian Splatting, including camera-motion blur, pose error, defocus, and color shifts. It represents motion blur as uncertainty over camera poses and adds image-specific depth and color corrections. On 49 ScanNet++ scenes, it improves SSIM and LPIPS over standard Gaussian Splatting, while Nerfacto has higher PSNR; motion modeling provides clearest ablation support.

Key insights

  • Integrating blur and color correction directly into the 3DGS optimization pipeline maintains computational efficiency while improving robustness.
  • Explicitly modeling motion blur as a distribution over camera poses and defocus blur via depth-dependent Gaussian covariance adjustments enables robust scene reconstruction from imperfect inputs.
  • Addressing color inconsistencies with per-image affine transformations significantly contributes to reconstruction quality on diverse real-world datasets.

Summary by alphaXiv.

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