Continuous Splatting meets Retinex: Continuous Gaussian Splatting and Implicit Reflectance Modeling for Low-Light Image Enhancement

2026-06-15 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

CGS-Retinex is introduced as a novel low-light image enhancement framework, marking the first to employ explicit-implicit joint modeling. This system integrates continuous Gaussian splatting with Retinex theory to address common issues like color distortion and structural artifacts in low-illumination images. It represents the image grid as a continuous parameter field, utilizing a continuous Gaussian renderer to estimate spatially continuous global illumination, thereby eliminating grid artifacts from discrete sampling. Furthermore, CGS-Retinex incorporates an implicit neural representation to model reflectance independently, guided by shallow high-frequency features for accurate texture detail reconstruction. Within the Retinex framework, it applies physics-inspired brightness consistency and illumination smoothness regularization to ensure proper exposure and high-fidelity recovery of structures and colors by precisely decoupling illumination and texture. Experiments show it suppresses dark-region noise and overexposure while restoring high-frequency details and colors.

Key takeaway

For Computer Vision Engineers developing low-light image enhancement systems, CGS-Retinex offers a robust approach to mitigate common issues like color distortion and noise. You should consider integrating continuous Gaussian splatting and implicit reflectance modeling to achieve superior decoupling of illumination and texture. This method promises enhanced structural fidelity and color restoration, potentially improving downstream vision task performance by providing clearer, artifact-free inputs.

Key insights

CGS-Retinex enhances low-light images by decoupling illumination and reflectance using continuous Gaussian splatting and implicit neural representation.

Principles

• Continuous parameter fields eliminate discrete sampling artifacts.
• Implicit neural representations can model reflectance independently.
• Physics-inspired constraints improve illumination and texture recovery.

Method

CGS-Retinex uses a continuous Gaussian renderer for global illumination and an implicit neural representation for reflectance, guided by high-frequency features, within a Retinex framework with physics-inspired constraints.

In practice

• Apply continuous splatting to avoid grid artifacts.
• Use implicit models for independent component representation.
• Integrate physics-inspired regularization for robust enhancement.

Topics

• Low-Light Image Enhancement
• Gaussian Splatting
• Retinex Theory
• Implicit Neural Representation
• Image Reconstruction
• Computer Vision

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• TurboGS: Accelerating 3D Gaussian Splatting via Error-Guided Sparse Pixel Sampling and Optimization — TurboGS accelerates 3DGS by focusing optimization on perceptually informative pixels using error-guided sampling and dynamic density control.
• Multinex: An ultra lightweight AI model advancing low light image enhancement — Multinex uses classical color vision and Retinex principles for efficient, compact low-light image enhancement.
• Voronoi-guided Bilateral 2D Gaussian Splatting for Arbitrary-Scale Hyperspectral Image Super-Resolution — GaussianHSI uses Voronoi-guided 2D Gaussian splatting for arbitrary-scale hyperspectral image super-resolution.
• PIXLRelight: Controllable Relighting via Intrinsic Conditioning — PIXLRelight enables physically accurate, controllable single-image relighting by bridging PBR and learned synthesis via intrinsic conditioning.
• IG-Diff: Complex Night Scene Restoration with Illumination-Guided Diffusion Model — Illumination-guided diffusion models effectively restore complex night scenes by integrating illumination priors and synthetic data.
• RPC-GS: Gaussian Splatting with native RPC Rendering for Satellite Imagery — Native RPC integration in Gaussian Splatting significantly improves satellite imagery 3D reconstruction accuracy.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.