VEDAL: Variational Error-Driven Asynchronous Learning for 3D Gaussian Splatting Pruning
2026-06-01 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors present VEDAL, a new method to reduce the memory needed by 3D Gaussian Splatting (3DGS), which is used to create realistic 3D views quickly but requires a lot of memory. Instead of previous guesses or batch updates, they use a smart mathematical approach that decides which parts to remove based on how uncertain the model is about them. This approach balances keeping good image quality with using fewer parts. Their tests show VEDAL compresses the model about five times smaller while only slightly reducing image quality and still rendering very fast.
3D Gaussian Splattingnovel view synthesispruningvariational free energyreconstruction uncertaintylatent variablesreal-time renderingPSNRMip-NeRF 360Tanks&Temples
Authors
Aoduo Li, Jiancheng Li, Huan Ye, Hongjian Xu, Shiting Wu, Xiujun Zhang, Zimeng Li, Xuhang Chen
Abstract
3D Gaussian Splatting (3DGS) achieves remarkable novel view synthesis quality with real-time rendering, yet suffers from excessive memory consumption due to millions of Gaussian primitives. Existing pruning methods rely on heuristic importance scores or synchronous batch updates, leading to suboptimal compression and training instability. We propose VEDAL, a principled framework that formulates Gaussian pruning as variational free energy minimization. Our approach introduces (1) a prediction-error gating mechanism that asynchronously activates pruning based on per-Gaussian reconstruction uncertainty, and (2) a variational uncertainty head that models pruning decisions as latent variables with learnable priors. The free energy objective naturally balances reconstruction fidelity against model complexity through an information-theoretic lens. Extensive experiments on Mip-NeRF 360, Tanks&Temples, and Deep Blending demonstrate that VEDAL achieves 5.2x compression with only 0.31 dB PSNR drop, outperforming PUP 3D-GS by +0.05 dB at a higher compression ratio and LightGaussian by +0.35 dB at comparable quality, while maintaining real-time rendering at 185 FPS.