Research map / Reinforcement learning / Visual and representation RL
RL-ScanIQA: Reinforcement-Learned Scanpaths for Blind 360deg Image Quality Assessment
Yujia Wang, Yuyan Li, Jiuming Liu, Fang-Lue Zhang, Xinhu Zheng, Neil.A Dodgson
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 |
|---|---|---|---|
| jointly train scanpath policy and assessor with RL instead of two-stage imitation training (ablation) | RL-ScanIQA · JUFE | SRCC: 0.651 -> 0.816 | Improved |
| use learned RL scanpaths instead of human ground-truth scanpaths (ablation) | RL-ScanIQA · JUFE | SRCC: 0.724 -> 0.816 | Improved |
| use learned RL scanpaths instead of human ground-truth scanpaths (ablation) | RL-ScanIQA · JUFE | PLCC: 0.752 -> 0.902 | Improved |
| add step-wise exploration rewards (ablation) | RL-ScanIQA · JUFE | SRCC: 0.754 -> 0.816 | Improved |
| add scanpath diversity-aware rewards (ablation) | RL-ScanIQA · JUFE | SRCC: 0.731 -> 0.816 | Improved |
| add task-aligned perceptual rewards (ablation) | RL-ScanIQA · CVIQD | SRCC: 0.921 -> 0.968 | Improved |
| add task-aligned perceptual rewards (ablation) | RL-ScanIQA · OIQA | SRCC: 0.874 -> 0.941 | Improved |
| add distortion-space augmentation with rank-consistent losses (ablation) | RL-ScanIQA · Cross-dataset, train CVIQD test OIQA | SRCC: 0.782 -> 0.901 | Improved |
| add distortion-space augmentation with rank-consistent losses (ablation) | RL-ScanIQA · Cross-dataset, train JUFE test CVIQD | SRCC: 0.664 -> 0.771 | Improved |
| increase number of scanpaths K beyond 15 at inference (ablation) | RL-ScanIQA · CVIQD, OIQA, JUFE | SRCC | No clear effect |
| increase scanpath length T from 7 to 15 (ablation) | RL-ScanIQA · CVIQD, OIQA, JUFE | SRCC | No clear effect |
| shorten scanpath length to T=4 (ablation) | RL-ScanIQA · CVIQD, OIQA, JUFE | SRCC | Hurt |
Closest papers by meaning
- Learned Scanpaths Aid Blind Panoramic Video Quality Assessment (CVPR 2024 · similarity 0.90)
- VisualQuality-R1: Reasoning-Induced Image Quality Assessment via Reinforcement Learning to Rank (NeurIPS 2025 · similarity 0.89)
- Refine-IQA: Multi-Stage Reinforcement Finetuning for Perceptual Image Quality Assessment (AAAI 2026 · similarity 0.88)
- Q-Insight: Understanding Image Quality via Visual Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Q-Probe: Scaling Image Quality Assessment to High Resolution via Context-Aware Agentic Probing (NeurIPS 2026 · similarity 0.88)
- Distilling Spatially-Heterogeneous Distortion Perception for Blind Image Quality Assessment (CVPR 2025 · similarity 0.87)
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.
