Research map / Image generation / Super-resolution
Real-World Image Super-Resolution as Multi-Task Learning
Wenlong Zhang, Xiaohui Li, Guangyuan SHI, Xiangyu Chen, Yu Qiao, Xiaoyun Zhang, Xiao-Ming Wu, Chao Dong
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 |
|---|---|---|---|
| fine-tune on identified task groups (TG) | RealESRGAN · DIV2K5G Group1 | PSNR: 20.10 -> 21.10 | Improved |
| fine-tune on identified task groups (TG) | RealESRGAN · DIV2K5G Group1 | LPIPS: 0.5355 -> 0.5056 | Improved |
| fine-tune on identified task groups (TG) | RealESRGAN · DIV2K5G Group0 | PSNR: 23.85 -> 23.99 | Improved |
| fine-tune on identified task groups (TG) | RealSwinIR · DIV2K5G Group1 | PSNR: 19.60 -> 20.62 | Improved |
| fine-tune on identified task groups (TG) | RealHAT · DIV2K5G Group1 | PSNR: 20.64 -> 21.41 | Improved |
| fine-tune on identified task groups (TG) | RealHAT · DIV2K5G Group0 | LPIPS: 0.4084 -> 0.4110 | Hurt |
| fine-tune on identified task groups (TG) | RealESRGAN · AIM2019 | PSNR: 23.89 -> 24.27 | Improved |
| fine-tune on identified task groups (TG) | RealESRGAN · RealSRset-Nikon | LPIPS: 0.3820 -> 0.3819 | No clear effect |
| fine-tune on identified task groups (TG) | RealESRGAN · DIV2K8G Group1 | PSNR: 19.92 -> 20.97 | Improved |
| use task grouping instead of random grouping (ablation) | RealESRGAN-TG · DIV2K validation groups | PSNR: about 0.2dB -> up to 1dB | Improved |
| use task grouping instead of random grouping (ablation) | RealESRGAN-TG · DIV2K validation groups | LPIPS: about 0.01 -> 0.03 | Improved |
| fine-tune on tasks with high versus low performance indicator (ablation) | RealESRGAN · single degradation tasks | PSNR: about 0.2 dB -> about 0.4-1.4dB | Improved |
Closest papers by meaning
- Beyond Image Super-Resolution for Image Recognition with Task-Driven Perceptual Loss (CVPR 2024 · similarity 0.88)
- Towards Realistic Data Generation for Real-World Super-Resolution (ICLR 2025 · similarity 0.87)
- LucidNFT: LR-Anchored Multi-Reward Preference Optimization for Flow-Based Real-World Super-Resolution (NeurIPS 2026 · similarity 0.86)
- Learning Coupled Dictionaries from Unpaired Data for Image Super-Resolution (CVPR 2024 · similarity 0.86)
- Pixel-level and Semantic-level Adjustable Super-resolution: A Dual-LoRA Approach (CVPR 2025 · similarity 0.86)
- DP²O-SR: Direct Perceptual Preference Optimization for Real-World Image Super-Resolution (NeurIPS 2025 · similarity 0.86)
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.
