Fourier Splatting: Generalized Fourier encoded primitives for scalable radiance fields

2026-03-20

AI Summary

Existing radiance field methods struggle to achieve continuous and efficient multi-scale rendering, as their control over detail relies on discrete pruning of primitive counts. This work proposes a scalable primitive representation based on Fourier-encoded descriptors, enabling dynamic level-of-detail adjustment within a single model by truncating Fourier coefficients. The formulation supports arbitrary closed shapes and is inherently scalable. By integrating a straight-through gradient estimator with the HYDRA densification strategy under an MCMC framework, the method achieves stable optimization and effective decomposition of complex primitives. Evaluated on standard benchmarks, it attains state-of-the-art rendering quality among planar-primitive approaches and matches leading voxel-based methods in perceptual metrics, making it well-suited for high-fidelity rendering under bandwidth constraints.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Search and Optimization: Sampling/Simulation-based Search
• Machine Learning: Learning with Manifolds

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Large-scale security measurements

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.

Problem

• scalable radiance fields
• novel view synthesis
• Fourier encoded primitives
• level of detail
• 3D Gaussian Splatting

Innovation

• Fourier Splatting
• scalable primitives
• Fourier encoded descriptors
• radiance fields
• level-of-detail rendering