Research map / Image generation / Super-resolution
FinPercep-RM: A Fine-grained Reward Model and Co-evolutionary Curriculum for RL-based Real-world Super-Resolution
Yidi Liu, Zihao Fan, Jie Huang, Jie Xiao, Dong Li, Wenlong Zhang, LEI BAI, Xueyang Fu, Zheng-jun Zha
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 |
|---|---|---|---|
| use CLIP-IQA as ReFL reward | DiT4SR · RealLQ250 | MUSIQ: 71.832 -> 70.456 | Hurt |
| use FinPercep-RM as ReFL reward instead of CLIP-IQA | DiT4SR · RealLQ250 | MUSIQ: 70.456 -> 73.456 | Improved |
| use FinPercep-RM reward | DiT4SR · DrealSR | LPIPS: 0.365 -> 0.342 | Improved |
| replace CLIP-IQA reward with FinPercep-RM under DPO | DiT4SR · RealLQ250 | MUSIQ: 69.234 -> 72.789 | Improved |
| replace CLIP-IQA reward with FinPercep-RM under GRPO | DiT4SR · RealLQ250 | MUSIQ: 68.567 -> 72.123 | Improved |
| add DINO feature-level difference to ground-truth degradation map (ablation) | FinPercep-RM · RealLQ250 | MUSIQ: 72.123 -> 73.456 | Improved |
| couple global score with degradation map instead of weighted sum (ablation) | FinPercep-RM · RealLQ250 | MUSIQ: 72.328 -> 73.456 | Improved |
| add CCL (parameter-incremental curriculum) (ablation) | FinPercep-RM · RealLQ250 | MUSIQ: 71.982 -> 73.456 | Improved |
| use parameter-incremental CCL instead of loss-weight curriculum (ablation) | FinPercep-RM · RealLQ250 | MUSIQ: 72.842 -> 73.456 | Improved |
| CLIP-IQA reward on CLIP-IQA metric | DiT4SR · DrealSR | ClipIQA: 0.548 -> 0.562 | Improved |
| FinPercep-RM on ClipIQA metric vs CLIP-IQA reward | SeeSR · DrealSR | ClipIQA: 0.567 -> 0.561 | No clear effect |
Closest papers by meaning
- DP²O-SR: Direct Perceptual Preference Optimization for Real-World Image Super-Resolution (NeurIPS 2025 · similarity 0.90)
- LucidNFT: LR-Anchored Multi-Reward Preference Optimization for Flow-Based Real-World Super-Resolution (NeurIPS 2026 · similarity 0.90)
- Augmenting Perceptual Super-Resolution via Image Quality Predictors (CVPR 2025 · similarity 0.88)
- Q-Insight: Understanding Image Quality via Visual Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- NSARM: Next-Scale Autoregressive Modeling for Robust Real-World Image Super-Resolution (NeurIPS 2026 · similarity 0.87)
- Chain-of-Zoom: Extreme Super-Resolution via Scale Autoregression and Preference Alignment (NeurIPS 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.
