CVPR2025
PUP 3D-GS: Principled Uncertainty Pruning for 3D Gaussian Splatting
Alex Hanson, Allen Tu, Vasu Singla, Mayuka Jayawardhana, Matthias Zwicker, Tom Goldstein
摘要
Figure 1 . We prune the 3D Gaussian Splatting (3D-GS) reconstruction of the Deep Blending playroom scene from 2.65M Gaussians to 0.265M Gaussians using our PUP 3D-GS pipeline, accelerating rendering speed from 76.65 FPS to 318.06 FPS -a 4.15× speed-up on this scene-while preserving fine details. In comparison, LightGaussian, a recent high-performing post-hoc pruning pipeline for pretrained 3D-GS models, loses substantially more fine details than PUP 3D-GS and achieves a slower rendering speed of 261.27 FPS.