Fourier Splatting: Generalized Fourier encoded primitives for scalable radiance fields

TL;DR

Fourier Splatting introduces scalable, Fourier-encoded primitives for radiance fields, enabling adjustable rendering quality and improved efficiency in novel view synthesis.

Contribution

It presents the first inherently scalable primitive for radiance field rendering using Fourier descriptors, allowing dynamic level-of-detail adjustment at runtime.

Findings

Achieves state-of-the-art quality among planar-primitive methods.

Provides comparable perceptual metrics to volumetric methods on benchmarks.

Enables rendering at varying levels of detail by truncating Fourier coefficients.

Abstract

Novel view synthesis has recently been revolutionized by 3D Gaussian Splatting (3DGS), which enables real-time rendering through explicit primitive rasterization. However, existing methods tie visual fidelity strictly to the number of primitives: quality downscaling is achieved only through pruning primitives. We propose the first inherently scalable primitive for radiance field rendering. Fourier Splatting employs scalable primitives with arbitrary closed shapes obtained by parameterizing planar surfels with Fourier encoded descriptors. This formulation allows a single trained model to be rendered at varying levels of detail simply by truncating Fourier coefficients at runtime. To facilitate stable optimization, we employ a straight-through estimator for gradient extension beyond the primitive boundary, and introduce HYDRA, a densification strategy that decomposes complex primitives into simpler constituents within the MCMC framework. Our method achieves state-of-the-art rendering quality among planar-primitive frameworks and comparable perceptual metrics compared to leading volumetric representations on standard benchmarks, providing a versatile solution for bandwidth-constrained high-fidelity rendering.