DARB-Splatting: Generalizing Splatting with Decaying Anisotropic Radial Basis Functions

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Pramuditha, Hashiru, Viruthshaan, Vinasirajan, Arunan, Vishagar, Nazar, Saeedha, Ramasinghe, Sameera, Lucey, Simon, Rodrigo, Ranga
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866914333393420288
author Pramuditha, Hashiru
Viruthshaan, Vinasirajan
Arunan, Vishagar
Nazar, Saeedha
Ramasinghe, Sameera
Lucey, Simon
Rodrigo, Ranga
author_facet Pramuditha, Hashiru
Viruthshaan, Vinasirajan
Arunan, Vishagar
Nazar, Saeedha
Ramasinghe, Sameera
Lucey, Simon
Rodrigo, Ranga
contents 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. With this fresh perspective, we demonstrate varying performances across selected DARB reconstruction kernels, achieving comparable training convergence and memory footprints, with on-par PSNR, SSIM, and LPIPS results.
format Preprint
id arxiv_https___arxiv_org_abs_2501_12369
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DARB-Splatting: Generalizing Splatting with Decaying Anisotropic Radial Basis Functions
Pramuditha, Hashiru
Viruthshaan, Vinasirajan
Arunan, Vishagar
Nazar, Saeedha
Ramasinghe, Sameera
Lucey, Simon
Rodrigo, Ranga
Computer Vision and Pattern Recognition
Artificial Intelligence
Graphics
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. With this fresh perspective, we demonstrate varying performances across selected DARB reconstruction kernels, achieving comparable training convergence and memory footprints, with on-par PSNR, SSIM, and LPIPS results.
title DARB-Splatting: Generalizing Splatting with Decaying Anisotropic Radial Basis Functions
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Graphics
url https://arxiv.org/abs/2501.12369