Luminance Residual Diffusion Network for Low-Light Image Enhancement

LumiRD model data addresses common low-light enhancement failures like noise amplification and structural degradation. The 871.5 MB archive contains model-related files, likely including weights and evaluation results, authored by Shengyu Fang and uploaded in April 2026.

Use Cases

• Training diffusion models for luminance-residual-guided restoration using the proposed Residual Diffusion Module (RDM).
• Benchmarking low-light enhancement methods against the Illumination Manifold Unfolding Module (IMUM) outputs for structural discriminability.
• Analyzing feature distributions regularized by the IMUM to improve robustness under complex illumination variations.
• Studying the integration of spatial, gradient, and frequency-domain cues for suppressed response unfolding in dark regions.

Strengths

• Model validated on multiple public low-light datasets for quantitative and visual performance.
• Archive size of 871.5 MB suggests inclusion of substantial model data or results.

Limitations

• Specific dataset row counts, image resolutions, and sample counts are not provided.
• The core data consists of model outputs and related files, not a primary image dataset for direct training.

Provenance

Source

Shengyu Fang via figshare.

Collection Method

Proposed as part of a research paper on the Luminance Residual Diffusion Network (LumiRD).

Time Range

null

Freshness

Last updated in April 2026.

Geography

null