![[2501.12369] DARB-Splatting: Generalizing Splatting with Decaying Anisotropic Radial Basis Functions](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2501.12369] DARB-Splatting: Generalizing Splatting with Decaying Anisotropic Radial Basis Functions
Summary
This paper introduces DARB-Splatting, a novel approach to 3D reconstruction using Decaying Anisotropic Radial Basis Functions, enhancing the efficiency and quality of splatting methods in computer vision.
Why It Matters
The research addresses limitations in current splatting techniques by exploring generalized reconstruction kernels, which could lead to improved performance in 3D visualizations and applications in graphics and AI, thereby broadening the scope of computer vision methodologies.
Key Takeaways
- DARB-Splatting generalizes splatting methods using new reconstruction kernels.
- The approach utilizes decaying anisotropic radial basis functions for better performance.
- Results show comparable training convergence and memory efficiency to existing methods.
- The study highlights the potential of non-exponential family functions in 3D reconstruction.
- Findings could influence future research and applications in computer vision and graphics.
Computer Science > Computer Vision and Pattern Recognition arXiv:2501.12369 (cs) [Submitted on 21 Jan 2025 (v1), last revised 17 Feb 2026 (this version, v3)] Title:DARB-Splatting: Generalizing Splatting with Decaying Anisotropic Radial Basis Functions Authors:Hashiru Pramuditha (1), Vinasirajan Viruthshaan (1), Vishagar Arunan (1), Saeedha Nazar (1), Sameera Ramasinghe (2), Simon Lucey (2), Ranga Rodrigo (1) ((1) University of Moratuwa, (2) University of Adelaide) View a PDF of the paper titled DARB-Splatting: Generalizing Splatting with Decaying Anisotropic Radial Basis Functions, by Hashiru Pramuditha (1) and 6 other authors View PDF Abstract:Splatting-based 3D reconstruction methods have gained popularity with the advent of 3D Gaussian Splatting, efficiently synthesizing high-quality novel views. These methods commonly resort to using exponential family functions, such as the Gaussian function, as reconstruction kernels due to their anisotropic nature, ease of projection, and differentiability in rasterization. However, the field remains restricted to variations within the exponential family, leaving generalized reconstruction kernels largely underexplored, partly due to the lack of easy integrability in 3D to 2D projections. In this light, we show that a class of decaying anisotropic radial basis functions (DARBFs), which are non-negative functions of the Mahalanobis distance, supports splatting by approximating the Gaussian function's closed-form integration advantage...