SVR-GS: Spatially Variant Regularization for Probabilistic Masks in 3D Gaussian Splatting
This work addresses the need for smaller, faster, and more memory-efficient 3D models for real-time applications like robotics and AR/VR, representing an incremental improvement over existing mask-based pruning methods.
The paper tackles the problem of reducing the number of Gaussians in 3D Gaussian Splatting for novel view synthesis by introducing a spatially variant regularizer that aligns sparsity pressure with per-pixel reconstruction loss, resulting in a 1.79× reduction in Gaussians compared to MaskGS and 5.63× compared to 3DGS with minimal PSNR drops of 0.50 dB and 0.40 dB, respectively.
3D Gaussian Splatting (3DGS) enables fast, high-quality novel view synthesis but typically relies on densification followed by pruning to optimize the number of Gaussians. Existing mask-based pruning, such as MaskGS, regularizes the global mean of the mask, which is misaligned with the local per-pixel (per-ray) reconstruction loss that determines image quality along individual camera rays. This paper introduces SVR-GS, a spatially variant regularizer that renders a per-pixel spatial mask from each Gaussian's effective contribution along the ray, thereby applying sparsity pressure where it matters: on low-importance Gaussians. We explore three spatial-mask aggregation strategies, implement them in CUDA, and conduct a gradient analysis to motivate our final design. Extensive experiments on Tanks\&Temples, Deep Blending, and Mip-NeRF360 datasets demonstrate that, on average across the three datasets, the proposed SVR-GS reduces the number of Gaussians by 1.79\(\times\) compared to MaskGS and 5.63\(\times\) compared to 3DGS, while incurring only 0.50 dB and 0.40 dB PSNR drops, respectively. These gains translate into significantly smaller, faster, and more memory-efficient models, making them well-suited for real-time applications such as robotics, AR/VR, and mobile perception.