DiffSoup: Direct Differentiable Rasterization of Triangle Soup for Extreme Radiance Field Simplification

2026-03-28

AI Summary

Existing radiance field methods struggle to simultaneously achieve high visual quality, efficient transmission, and real-time rendering. To address this challenge, this work proposes an extremely compact radiance field representation composed of a sparse “triangle soup,” where each triangle carries neural textures and binary opacity values, enabling end-to-end training through a standard rasterization pipeline. The key innovation lies in the first differentiable rasterizer capable of directly handling binary-opacity triangle soups without requiring opacity smoothing, thereby supporting depth testing and seamless integration with conventional graphics pipelines. Differentiability is achieved via stochastic opacity masking during rendering. The resulting method enables interactive frame rates on consumer-grade laptops and mobile devices while compressing model size by several orders of magnitude, all without compromising visual fidelity.

Abstract

Radiance field reconstruction aims to recover high-quality 3D representations from multi-view RGB images. Recent advances, such as 3D Gaussian splatting, enable real-time rendering with high visual fidelity on sufficiently powerful graphics hardware. However, efficient online transmission and rendering across diverse platforms requires drastic model simplification, reducing the number of primitives by several orders of magnitude. We introduce DiffSoup, a radiance field representation that employs a soup (i.e., a highly unstructured set) of a small number of triangles with neural textures and binary opacity. We show that this binary opacity representation is directly differentiable via stochastic opacity masking, enabling stable training without a mollifier (i.e., smooth rasterization). DiffSoup can be rasterized using standard depth testing, enabling seamless integration into traditional graphics pipelines and interactive rendering on consumer-grade laptops and mobile devices. Code is available at https://github.com/kenji-tojo/diffsoup.

Problem

• radiance field simplification
• efficient rendering
• cross-platform transmission
• real-time rendering
• model compression

Innovation

• differentiable rasterization
• triangle soup
• radiance field simplification
• neural textures
• binary opacity