New model LEIQ-Assessor evaluates low-light image enhancement quality

Researchers have developed LEIQ-Assessor, a novel multi-task learning model designed to evaluate the quality of low-light enhanced images. Utilizing a SigLIP2 Vision Transformer backbone, the model simultaneously predicts the overall Mean Opinion Score (MOS) and six specific perceptual attributes, including lightness, color fidelity, and content recovery. This approach captures richer quality-aware features compared to single-task models. LEIQ-Assessor demonstrated superior performance on the MLE benchmark, outperforming existing no-reference IQA models and achieving second place in the QoMEX 2026 Grand Challenge on Low-light Enhanced Image Quality Assessment. AI

IMPACT This model could improve the development and practical application of low-light image enhancement algorithms by providing a more comprehensive quality assessment.

RANK_REASON This is a research paper describing a new model for image quality assessment. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• LEIQ-Assessor
• MLE benchmark
• QoMEX 2026 Grand Challenge on Low-light Enhanced Image Quality Assessment
• SigLIP2 Vision Transformer

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →