Research map / Reinforcement learning / Visual and representation RL
TIR-Agent: Training an Explorative and Efficient Agent for Image Restoration
Yisheng Zhang, Guoli Jia, Haote Hu, Shanxu Zhao, Kaikai Zhao, Long Sun, Xinwei Long, Kai Tian, Che Jiang, Zhaoxiang Liu, Kai Wang, Shiguo Lian and 2 more
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 SFT stage before RL (ablation) | TIR-Agent · MiO100 Group A | PSNR: 19.49 -> 22.07 | Improved |
| add RL stage after SFT (ablation) | TIR-Agent · MiO100 Group A | MUSIQ: 60.17 -> 63.69 | Improved |
| add exploration-driven perturbation (EDP) to SFT data (ablation) | TIR-Agent · MiO100 Group A | PSNR: 19.97 -> 22.07 | Improved |
| add task-order perturbation alpha^t (ablation) | TIR-Agent · MiO100 Group A | PSNR: 20.61 -> 22.07 | Improved |
| add tool-level perturbation alpha^m (ablation) | TIR-Agent · MiO100 Group A | LPIPS: 0.3256 -> 0.2874 | Improved |
| add multi-dimensional adaptive reward (MAR) (ablation) | TIR-Agent · MiO100 Group A | PSNR: 20.85 -> 22.07 | Improved |
| add MAR (ablation) | TIR-Agent · MiO100 Group A | LPIPS: 0.2830 -> 0.2874 | Hurt |
| add decoupled per-metric advantage normalization (ablation) | TIR-Agent · MiO100 Group A | PSNR: 21.86 -> 22.07 | Improved |
| add adaptive metric weights omega^theta (ablation) | TIR-Agent · MiO100 Group A | PSNR: 20.51 -> 22.07 | Improved |
| increase task-order perturbation probability alpha^t above 0.3 (ablation) | TIR-Agent · MiO100 Group A | PSNR | Hurt |
| increase tool perturbation alpha^m (ablation) | TIR-Agent · MiO100 Group A | PSNR | No clear effect |
| train agent with SFT+RL instead of training-free tool traversal | TIR-Agent vs AgenticIR · MiO100 Group A | Inference time (s): 184.57 -> 61.58 | Improved |
Closest papers by meaning
- OPERA: An Agent for Image Restoration with End-to-End Joint Planning–Execution Optimization (NeurIPS 2026 · similarity 0.93)
- An Intelligent Agentic System for Complex Image Restoration Problems (ICLR 2025 · similarity 0.90)
- Hybrid Agents for Image Restoration (CVPR 2026 · similarity 0.90)
- RestoreAgent: Autonomous Image Restoration Agent via Multimodal Large Language Models (NeurIPS 2024 · similarity 0.89)
- Complexity Experts are Task-Discriminative Learners for Any Image Restoration (CVPR 2025 · similarity 0.88)
- Continuous Expert Assembly: Instance-Conditioned Low-Rank Residuals for All-in-One Image Restoration (NeurIPS 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.
