Research map / Image generation / Super-resolution
OARS: Process-Aware Online Alignment for Generative Real-World Image Super-Resolution
Shijie Zhao, Xuanyu Zhang, Bin Chen, Weiqi Li, Qunliang Xing, Kexin Zhang, Yan Wang, Junlin Li, Li Zhang, Jian Zhang, Tianfan Xue
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 three-stage score calibration (ablation) | COMPASS reward · COMPASS-20K reward ablation | accuracy: 78.8 -> 81.5 | Improved |
| add explicit fidelity modeling (ablation) | COMPASS reward · COMPASS-20K reward ablation | accuracy: 81.5 -> 82.3 | Improved |
| add quality-adaptive gamma=7 (ablation) | COMPASS reward · COMPASS-20K reward ablation | accuracy: 82.3 -> 83.1 | Improved |
| vary gamma among 5, 7, 9 (ablation) | COMPASS reward · COMPASS-20K reward ablation | accuracy: 82.7 / 83.1 / 82.8 | No clear effect |
| run RL on base model weights instead of SFT weights (stage 1) (ablation) | OARS · RealSR | LIQE: 4.0784 -> 4.2345 | Improved |
| run RL on base model weights instead of SFT weights (both stages) (ablation) | OARS · RealSR | PSNR: 21.31 -> 22.36 | Improved |
| add stage-2 reference-free RL (ablation) | OARS on base · RealSR | LIQE: 4.2345 -> 4.3045 | Improved |
| apply OARS to Qwen-SFT | Qwen-Image-Edit-2509 · RealSR | LIQE and PSNR: LIQE 3.8146 -> 4.3045; PSNR 22.71 -> 22.36 | Mixed |
| use COMPASS reward instead of RALI (ablation) | OARS RL · RealSR | PSNR: 20.83 -> 22.36 | Improved |
| add fidelity gating term to perceptual-gain-only reward (ablation) | COMPASS reward · RealSR | PSNR: 21.38 -> 21.89 | Improved |
| add input-quality-aware fidelity term to gated reward (ablation) | COMPASS reward · RealSR | PSNR: 21.89 -> 22.36 | Improved |
| use NFT-based OARS instead of Flow-GRPO (ablation) | COMPASS reward · RealSR | LIQE: 3.9351 -> 4.3045 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SeeSR — SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024) · better on 0 of 1 shared result
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
- LucidNFT: LR-Anchored Multi-Reward Preference Optimization for Flow-Based Real-World Super-Resolution (NeurIPS 2026 · similarity 0.88)
- Chain-of-Zoom: Extreme Super-Resolution via Scale Autoregression and Preference Alignment (NeurIPS 2025 · similarity 0.88)
- DP²O-SR: Direct Perceptual Preference Optimization for Real-World Image Super-Resolution (NeurIPS 2025 · similarity 0.87)
- NSARM: Next-Scale Autoregressive Modeling for Robust Real-World Image Super-Resolution (NeurIPS 2026 · similarity 0.86)
- SeeClear: Semantic Distillation Enhances Pixel Condensation for Video Super-Resolution (NeurIPS 2024 · similarity 0.85)
- SRSR: Enhancing Semantic Accuracy in Real-World Image Super-Resolution with Spatially Re-Focused Text-Conditioning (NeurIPS 2025 · similarity 0.85)
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.
