Geometry-aware Depth-guided Representation Learning for Structure-preserving Low-light Image Enhancement

2026-07-06

AI Summary

This work addresses the challenge of structural information loss in low-light image enhancement, a problem often exacerbated by existing methods that overlook the critical role of scene geometry in preserving structural consistency. To this end, we propose Depth-guided Multi-scale Attention Network (DMSA-Net), which, for the first time, integrates depth priors into low-light representation learning. By jointly leveraging Retinex decomposition and depth estimation within a hierarchical encoder-decoder architecture, our approach enables deep fusion of geometric and appearance features. Key contributions include a novel depth-guided multi-scale fusion strategy and LOL-D, the first real-world low-light dataset annotated with ground-truth depth maps. Extensive experiments demonstrate that our method significantly improves structural fidelity and visual quality across multiple benchmarks, advancing the frontier of geometry-aware low-light enhancement.

Abstract

Low-light degradation reduces image visibility and weakens structural cues that are important for visual representation and scene understanding. Existing low-light image enhancement methods mainly focus on appearance restoration, while insufficiently exploiting scene geometry to preserve structural consistency. To address this limitation, this paper proposes a Depth-guided Multi-scale Attention Network (DMSA-Net) for geometry-aware low-light image enhancement. DMSA-Net introduces depth-related structural priors into low-light representation learning through reflectance-geometry interaction. A Retinex-based decomposition module is first used to obtain illumination-invariant reflectance representations, from which depth cues are inferred to characterize scene structure under degraded illumination. A multi-scale depth-guided fusion strategy is then embedded into a hierarchical encoder-decoder architecture, where depth-aware attention adaptively integrates geometric and appearance features. Experiments on several benchmark datasets show that DMSA-Net achieves effective low-light restoration while improving structural preservation. Moreover, we construct LOL-D, a depth-augmented low-light dataset, to facilitate research on geometry-aware low-light vision.

Problem

• low-light image enhancement
• structural preservation
• scene geometry
• depth cues
• geometry-aware representation

Innovation

• depth-guided
• geometry-aware
• low-light enhancement
• multi-scale attention
• structural preservation