Research map / Vision-language and multimodal / Multimodal LLMs
ME-IQA: Memory-Enhanced Image Quality Assessment via Re-Ranking
Kanglong Fan, Tianhe Wu, Wen Wen, Jianzhao Liu, le yang, Yabin ZHANG, Yiting Liao, Junlin Li, Li Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ME-IQA test-time memory re-ranking | VisualQuality-R1 · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.698 -> 0.726 | Improved |
| add ME-IQA test-time memory re-ranking | EvoQuality · IQA WAVG over 7 benchmarks (KonIQ-trained) | SRCC: 0.716 -> 0.751 | Improved |
| add ME-IQA test-time memory re-ranking | GPT-5 · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.676 -> 0.706 | Improved |
| replace 64-sample majority voting with ME-IQA@32 | VisualQuality-R1 · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.710 -> 0.726 | Improved |
| increase neighborhood size K from 0 to 32 (ablation) | ME-IQA · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.698 -> 0.726 | Improved |
| increase K from 32 to 256 (ablation) | ME-IQA · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.726 -> 0.735 | No clear effect |
| lengthen direct reasoning to L=500 instead of ME-IQA (ablation) | Direct scoring · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.698 -> 0.689 | Hurt |
| use reasoning embeddings instead of random retrieval (ablation) | ME-IQA · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.703 -> 0.726 | Improved |
| use reasoning embeddings instead of image embeddings (ablation) | ME-IQA · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.706 -> 0.726 | Improved |
| add consistency reward in RL finetuning (reasoning embeddings*) (ablation) | ME-IQA · IQA WAVG over 7 benchmarks (KonIQ-trained) | PLCC: 0.726 -> 0.737 | Improved |
| use pairwise Thurstone re-ranking instead of kernel re-scoring (ablation) | ME-IQA · KADID | PLCC: 0.707 -> 0.741 | Improved |
| add quadratic prior lambda=1e-2 to Thurstone fusion (ablation) | ME-IQA · KADID | PLCC: 0.724 -> 0.741 | Improved |
Closest papers by meaning
- Revisiting MLLM Based Image Quality Assessment: Errors and Remedy (AAAI 2026 · similarity 0.91)
- Reasoning as Representation: Rethinking Visual Reinforcement Learning in Image Quality Assessment (ICLR 2026 · similarity 0.90)
- Self-Evolving Vision-Language Models for Image Quality Assessment via Voting and Ranking (ICLR 2026 · similarity 0.90)
- VisualQuality-R1: Reasoning-Induced Image Quality Assessment via Reinforcement Learning to Rank (NeurIPS 2025 · similarity 0.89)
- IQA-Spider: Unifying Multi-Granularity Image Quality Assessment with Reasoning, Grounding and Referring (ICML 2026 · similarity 0.89)
- Grounding-IQA: Grounding Multimodal Language Model for Image Quality Assessment (ICLR 2026 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
