PCAsplat: differentiable local PCA regularization aligns Gaussians to surfaces and kills floaters
zhenjun_zhao · x · 2026-10-09
PCAsplat, from a team including Stanford's Gordon Wetzstein and Leonidas Guibas, tackles a Gaussian Splatting weakness: rasterization losses leave occluded or low-contribution Gaussians with weak gradients, causing floaters. It introduces a differentiable local-PCA regularizer acting on Gaussian-center neighborhoods so even view-irrelevant Gaussians get updated; eigenvalue regularization pushes Gaussians onto the surface with isotropic tangent-plane coverage, and normals are aligned to PCA estimates. On DTU, Tanks and Temples and NeRF Synthetic it yields surface-aligned splats, far fewer floaters, and enables point-cloud segmentation and direct Poisson reconstruction.
More from Research
- Embedding arithmetic demo: cartoon − animation − comic = diagram — mayfer · 2026-10-23
- Domingos Cites New Result Claiming RL Is 'Basically Useless' vs Supervised Learning — pmddomingos · 2026-10-23
- Wolfram Opens Rule 30 Prizes to AIs: $30,000 for Three Open Problems — cephaloform · 2026-10-23
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23
- EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23