SplatQ improves 3D Gaussian pruning for better scene balance and speed
SplatQ: Interaction-Aware Pruning of 3D Gaussian Splatting via Quadratic Selection
Computer Vision and Pattern RecognitionGraphics
Summary
3D scenes often use many small glowing dots called Gaussians to show colors and shapes. Old methods remove these dots one by one based only on their importance, which can cause uneven picture quality. The authors created SplatQ, a smarter way to remove dots that looks at groups together and how they affect the picture when removed as a team. This method balances the dots across the scene and speeds up drawing while keeping the picture sharp. Tests show SplatQ works better than previous methods in keeping details and quality evenly across the scene.
What this means in practice
- •For 3d graphics developers: Optimize 3D Gaussian models in real-time rendering systems to balance detail quality while reducing computational load.
- •For game engine designers: Improve scene rendering speed and visual consistency by applying coordinated pruning of Gaussian splats with budget controls.
Authors
Hongbin Ye, Haixu Liu, Shenwei Kang, Wenxiu Wei, Tinglan Wang, Hongdong Zhu, Fan Tang, Qi Gao, Yin Ma, Kai Wen, Hai Wei, Nuo Qian, Wei-Shi Zheng, Shaobo Chen, Fa-Ting Hong, Jingke Meng
Abstract
Existing 3D Gaussian Splatting (3DGS) pruning methods primarily rely on global Top-K selection based on independent importance scores, lacking regional budget control and ignoring joint deletion effects. Consequently, they cannot ensure balanced reconstruction quality across a scene. We propose SplatQ, a combinatorial pruning algorithm that coordinates regional budget allocation and joint Gaussian selection under a fixed budget. SplatQ partitions Gaussians by spatial proximity and appearance similarity into size-limited blocks, decomposing global selection into subproblems of a specific size. It then smooths importance-based block retention quotas to prevent excessive budget concentration. Finally, it constructs a quadratic approximation of joint deletion effects using the responses of representative pixel colors to changes in Gaussian opacity. Individual responses, pairwise interactions, and quota penalties are incorporated into a quadratic unconstrained binary optimization (QUBO) objective, solved using simulated annealing (SA) or a coherent Ising machine (CIM) to jointly select Gaussians. Experiments on 5 benchmarks demonstrate that SplatQ outperforms state-of-the-art pruning methods in aggregate reconstruction quality, spatial quality balance, and rendering speed. Ablations confirm that budget smoothing improves local detail reconstruction, while interaction modeling improves Gaussian selection quality under a fixed budget.