ME-IQA: Memory-Enhanced Image Quality Assessment via Re-Ranking

2026-03-21

AI Summary

This work addresses the limitation of existing reasoning-driven vision-language models (VLMs) in image quality assessment (IQA), which often suffer from discrete collapse due to reliance on scalar scores and struggle to capture fine-grained distortion differences. To overcome this, the authors propose a plug-and-play, test-time memory-augmented reranking framework. It dynamically maintains a memory bank of reference samples and retrieves semantically and perceptually aligned neighbors using reasoning summaries. The VLM is reformulated as a probabilistic comparator grounded in the Thurstone Case V psychometric model, integrating ordinal pairwise preference evidence to refine initial quality predictions. This approach, the first to incorporate test-time memory mechanisms with probabilistic preference modeling, significantly outperforms current VLMs, traditional IQA methods, and test-time scaling strategies across multiple benchmarks, yielding denser and more distortion-sensitive quality estimates.

Technology Category

• Computer Vision: Large Vision Models
• Knowledge Representation and Reasoning: Qualitative Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Reasoning-induced vision-language models (VLMs) advance image quality assessment (IQA) with textual reasoning, yet their scalar scores often lack sensitivity and collapse to a few values, so-called discrete collapse. We introduce ME-IQA, a plug-and-play, test-time memory-enhanced re-ranking framework. It (i) builds a memory bank and retrieves semantically and perceptually aligned neighbors using reasoning summaries, (ii) reframes the VLM as a probabilistic comparator to obtain pairwise preference probabilities and fuse this ordinal evidence with the initial score under Thurstone's Case V model, and (iii) performs gated reflection and consolidates memory to improve future decisions. This yields denser, distortion-sensitive predictions and mitigates discrete collapse. Experiments across multiple IQA benchmarks show consistent gains over strong reasoning-induced VLM baselines, existing non-reasoning IQA methods, and test-time scaling alternatives.

Problem

• image quality assessment
• discrete collapse
• vision-language models
• scalar scores
• sensitivity

Innovation

• memory-enhanced
• re-ranking
• vision-language model
• image quality assessment
• discrete collapse