Research map / Vision-language and multimodal / Multimodal LLMs
Zoom-IQA: Image Quality Assessment with Reliable Region-Aware Reasoning
Guoqiang Liang, Jianyi Wang, Zhonghua Wu, Shangchen Zhou, Chen Change Loy
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 KL-Coverage regularizer (ablation) | Zoom-IQA RL · KonIQ (trained on KonIQ) | PLCC: 0.908 -> 0.932 | Improved |
| add KL-Coverage regularizer (ablation) | Zoom-IQA RL · CSIQ (trained on KonIQ) | PLCC: 0.759 -> 0.791 | Improved |
| add progressive re-sampling training (ablation) | Zoom-IQA · KonIQ (trained on KonIQ) | PLCC: 0.932 -> 0.938 | Improved |
| add progressive re-sampling training (ablation) | Zoom-IQA · CSIQ (trained on KonIQ) | PLCC: 0.791 -> 0.797 | Improved |
| use rank reward only versus score reward only (ablation) | Zoom-IQA RL · KADID (trained on KonIQ) | PLCC: 0.677 -> 0.709 | Mixed |
| use score reward only versus rank reward only (ablation) | Zoom-IQA RL · KonIQ (trained on KonIQ) | PLCC: 0.906 -> 0.928 | Mixed |
| add score reward to rank reward (ablation) | Zoom-IQA RL · KonIQ (trained on KonIQ) | PLCC: 0.906 -> 0.932 | Improved |
| add RL stage after SFT (ablation) | Zoom-IQA · KonIQ (trained on KonIQ) | PLCC: 0.836 -> 0.938 | Improved |
| use learned zoom policy instead of central crop (ablation) | Zoom-IQA · KonIQ (trained on KonIQ) | PLCC: 0.898 -> 0.938 | Improved |
| use two-stage SFT+RL instead of single-stage SFT (ablation) | Zoom-IQA · KonIQ (trained on KonIQ) | SRCC: 0.812 -> 0.922 | Improved |
| use Zoom-IQA reasoning instead of LLaVA-13b captions to guide SUPIR | SUPIR · DIV2K | MANIQA: 0.4974 -> 0.5455 | Improved |
| use Zoom-IQA reasoning instead of LLaVA-13b captions to guide SUPIR | SUPIR · DIV2K | PSNR: 21.57 -> 21.31 | Hurt |
Later papers that compared with this method
1 later accepted paper reports Zoom-IQA as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- IQA-T1: Tool-based Visual Evidence Reasoning for Image Quality Assessment (ECCV 2026 · better on 3 of 3 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Q-Insight — Q-Insight: Understanding Image Quality via Visual Reinforcement Learning (NeurIPS 2025) · better on 5 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- VisualQuality-R1: Reasoning-Induced Image Quality Assessment via Reinforcement Learning to Rank (NeurIPS 2025 · similarity 0.92)
- Reasoning as Representation: Rethinking Visual Reinforcement Learning in Image Quality Assessment (ICLR 2026 · similarity 0.92)
- Revisiting MLLM Based Image Quality Assessment: Errors and Remedy (AAAI 2026 · similarity 0.91)
- IQA-Spider: Unifying Multi-Granularity Image Quality Assessment with Reasoning, Grounding and Referring (ICML 2026 · similarity 0.91)
- Refine-IQA: Multi-Stage Reinforcement Finetuning for Perceptual Image Quality Assessment (AAAI 2026 · similarity 0.90)
- Self-Evolving Vision-Language Models for Image Quality Assessment via Voting and Ranking (ICLR 2026 · similarity 0.90)
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.
