EvenSplat: Coupled 2D-3D Decomposition for Gaussian Splatting under Exposure and Illumination Variation

2026-10-01

AI Summary

This study addresses the challenge of illumination-geometry color coupling and resulting artifacts caused by uneven exposure in multi-view reconstruction. To this end, we propose an illumination decoupling framework based on joint 2D-3D decomposition. This method introduces a novel shared illumination representation mechanism between image space and 3D Gaussian fields. By integrating a camera response network, a local exposure compensation module, and a joint optimization algorithm, it achieves cross-dimensionally consistent separation of scene attributes. Experimental results demonstrate that our approach significantly eliminates reconstruction artifacts under high-contrast lighting conditions across multiple datasets. It outperforms existing state-of-the-art methods, substantially improving the overall quality of 3D reconstruction.

Abstract

A surface photographed under even light presents nearly the same appearance from every angle; the same surface under uneven light does not. Exposure changes between views, illumination varies within a single image, and locally strong light sources leave one region bright and its neighbor in shadow. Multi-view reconstruction methods such as 3D Gaussian Splatting treat these lighting artifacts as if they were properties of the scene, entangling capture-specific illumination with the geometry and color they recover. We present EvenSplat, a framework that separates the two. EvenSplat couples an image-space illumination decomposition with an illumination field carried by the Gaussians, so that the same explanation of the lighting is shared between the two-dimensional and three-dimensional views of the scene; a camera-response network and a local exposure-compensation module absorb the global and residual differences that remain across training images. Through extensive experiments across multiple datasets and diverse forms of uneven illumination (cross-view exposure, spatial illumination variation, and high-contrast lighting) on both real-world captured and simulated benchmarks, EvenSplat generally outperforms state-of-the-art methods, particularly under high-contrast illumination.

Problem

• Gaussian Splatting
• illumination variation
• exposure changes
• multi-view reconstruction
• lighting disentanglement

Innovation

• 3D Gaussian Splatting
• Illumination Decomposition
• Exposure Compensation
• 2D-3D Coupling
• Camera Response Network