Reproduce “SG-Splatting: Accelerating 3D Gaussian Splatting with Spherical Gaussians” 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
Yiwen Wang, Siyuan Chen, Ran Yi
SG-Splatting reduces the storage and rendering cost of view-dependent color in 3D Gaussian Splatting by replacing third-degree spherical harmonics with compact spherical-Gaussian lobes, constraining three lobe orientations to be orthogonal, and adding adaptive low-degree harmonics. On Mip-NeRF360, it renders at 334 FPS and uses 416.02 MB, versus 247 FPS and 781.46 MB for 3D-GS, with near-baseline quality on RTX 3090.
Key insights
- Spherical Gaussians (SGs) offer a more compact and flexible parameterization (10 parameters vs. 48 for SH) for view-dependent color, efficiently capturing directional light.
- Enforcing orthogonality between multiple SGs effectively distributes captured scene content, reducing redundancy and enhancing expressiveness for complex lighting.
- A mixed representation combining adaptive low-degree spherical harmonics for diffuse components and SGs for high-frequency details achieves superior efficiency without significant quality degradation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.