A new benchmark called TrustUIQA has been developed to evaluate underwater image quality assessment metrics, as current metrics do not consistently align with human perception. Researchers found that even the best-performing metrics had low agreement with human judgments, and common metrics like UCIQE often favored edits that humans disliked. The benchmark includes human preference data, edited images, and adversarial perturbations to test metric robustness, revealing significant discrepancies between automated assessments and human visual preferences. AI
IMPACT Highlights a critical gap in automated image quality assessment, potentially impacting AI development for image enhancement and analysis.
RANK_REASON The item is an academic paper introducing a new benchmark for evaluating image quality metrics. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →