SPACE introduces a Depth-Adaptive HVI Transformation to decouple luminance and chrominance under depth guidance, effectively suppressing color-space noise and a Depth-Manifold Modulated Attention mechanism constrains feature interactions within a learned depth manifold, ensuring structural coherence during enhancement.
This work introduces See in the Degraded Extremely Dark (SIDED), a new dataset that applies controlled motion degradation to extremely low-light RAW pairs while retaining their original sensor noise and introduces a physics-guided refinement model to strengthen illumination--reflectance consistency, pixel fidelity, and color preservation without incurring additional inference cost.
Zepu Wang, Jin Liang, Weijie Xiao et al.· 0 citations
Recent low-light image enhancement (LLIE) methods have driven brightness and structural fidelity close to that of normally-exposed images, yet their outputs still exhibit systematic color shifts such as greenish skies, yellowish faces, and warm-tinted white objects. We attribute this to end-to-end LLIE training coupling brightness, structure, and color within a single network, leaving the color channels weakly supervised. We recast color restoration as an independent sub-problem and decouple it from brightness enhancement, realizing it as a general-purpose post-processing module built on retrieval-augmented generation (RAG). Rather than relying solely on parametric color priors learned during training, the module dynamically retrieves a reference image from an external high-quality color knowledge base and injects its color distribution into a color-restoration network to correct residual bias. The design has three components: (i) a dual-index FAISS retriever built on intermediate VGG19 features, capturing textural and structural similarity through global mean and variance statistics; (ii) GlobalSPHistAdaIN, which reduces the reference spatial-preserving color histogram to a global color vector and modulates network features via adaptive instance normalization, removing dependence on pixel-level correspondence; and (iii) a residual formulation that predicts a color correction over the front-end output. Across LOLv1, LOLv2-Real, and LOLv2-Synthetic, the module consistently improves color-specific metrics, and it remains effective when the front end is swapped among CPGA-Net++, LLFormer, FLIGHTNet, and IAT, confirming cross-front-end generality. Ablations show that a VGG19 dual index outperforms CLIP-based retrieval, indicating that color restoration depends on textural and structural similarity rather than high-level semantics.
Reconstructing 3D scenes under real-world low-light conditions remains challenging due to severe sensor noise, low signal-to-noise ratios, and degraded photometric consistency, which destabilize geometry estimation and novel view synthesis. Existing approaches often rely on well-lit reference data for reliable Structure-from-Motion (SfM) initialization under degraded inputs or apply per-view enhancement methods that introduce cross-view inconsistencies. To address these limitations, we propose \textbf{NOVA-GS}, a unified noise-aware framework for low-light 3D Gaussian Splatting that subsumes enhancement, denoising, and geometry optimization within a single process. Our method leverages VGGT-based feed-forward estimation to obtain robust camera poses and geometry directly from degraded inputs, eliminating the need for SfM. Building on this initialization, NOVA-GS integrates three coupled components: a structure-aware enhancement module for exposure correction, a self-supervised denoising module with blind-spot masking for pseudo-supervision, and a consistency-driven Gaussian Splatting optimization enforcing cross-view geometric coherence. We further introduce a noise-guided spherical harmonic regularization to suppress view-dependent artifacts in noisy regions. Extensive experiments on diverse real-world low-light datasets demonstrate improved geometric fidelity, color consistency, and robustness without requiring paired supervision or well-lit references. https://shaurya2524.github.io/nova-gs/
A. ShauryaPavan, Vemunuri Divya Madhuri, Yash Pradeep Gawande et al.· 0 citations
Low-light light field (L3F) images suffer from severe structural degradation, including low contrast, blurred edges, and heavy noise, which disrupts angular consistency. Existing single-image enhancement methods fail to exploit the spatial-angular consistency of light field (LF) images, while L3F enhancement methods struggle in low-light scenarios with extremely low contrast, often resulting in over-smoothed edges and the loss of geometric details. To address these issues, we propose the Edge-guided Hybrid Enhancement Network (EHENet), an efficient network integrating structural priors and spatial-angular extraction. We propose a structural prior embedding strategy that employs Scharr operators and Gaussian filtering to explicitly model spatial edges. Furthermore, we design the Global Feature Enhancement (GFE) block to extract spatial-angular correlation. Experiments show EHENet significantly outperforms state-of-the-art methods in both quality and efficiency.
Hao Wu, Bing-Jie Zhu, Shizheng Li et al.· International Conference on...· 0 citations
A Relative Illumination Structure Estimation (RISE) framework is proposed that decouples relative illumination structure from absolute exposure and infers it from reliable bright regions, enabling interpretable and robust enhancement.
Tian-Le Du, Peiyuan He, Hainuo Wang et al.· 0 citations
Low-light image enhancement is commonly formulated as an illumination recovery problem. However, severe illumination degradation not only reduces image brightness but also weakens structural responses, causing edge discontinuities and unstable texture reconstruction. To address these challenges, low-light enhancement is reconsidered as a structure-constrained reconstruction problem, where illumination recovery and structural continuity preservation are jointly optimized. Although supervised methods can achieve promising enhancement quality, their reliance on paired low-light and normal-light images limits their applicability to diverse real-world scenarios. Unsupervised approaches provide a more flexible solution by avoiding the requirement for paired training data. In this work, a continuity-guided low-light enhancement network, termed CGLEN, is proposed for unsupervised structure-aware image reconstruction. CGLEN introduces a learnable Retinex decomposition module to estimate illumination and reflectance components, followed by a gradient-guided structural representation that provides reliable structural cues during enhancement. Furthermore, a PDE-inspired structural continuity refinement strategy is developed by incorporating gradient variation and Laplacian consistency into a lightweight residual propagation framework, enabling spatial continuity preservation under degraded illumination conditions. A structure-guided modulation mechanism, together with an auxiliary reconstruction branch and adaptive fusion strategy, is further introduced to improve optimization stability and reconstruction consistency. Extensive experiments on multiple benchmark datasets demonstrate that CGLEN achieves competitive enhancement performance compared with existing supervised and unsupervised methods, while maintaining relatively low computational complexity. The results indicate that explicitly modeling structural continuity provides an effective strategy for unsupervised low-light image enhancement, particularly in challenging illumination conditions.
Yang Li, Ruobo Xu, Kai Zhou· Scientific Reports· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
