Research map / Image generation / Super-resolution
Crafting Training Degradation Distribution for the Accuracy-Generalization Trade-off in Real-World Super-Resolution
Ruofan Zhang, Jinjin Gu, Haoyu Chen, Chao Dong, Yulun Zhang, Wenming Yang
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 |
|---|---|---|---|
| reference-based training degradation distribution crafting (Fréchet-distance binning) vs uniform sampling over bins (ablation) | SRResNet · Synthetic settings ①-④ | PSNR: +0.71 (setting ①), +0.21 (setting ④) | Improved |
| reference-based training degradation distribution crafting vs uniform sampling over bins (ablation) | RRDB · Synthetic setting ① | PSNR: +0.48 | Improved |
| training with large-range complex degradations (Real-ESRNet/BSRNet) vs narrow crafted distribution | SRResNet-based SR · Synthetic setting ① | PSNR: -0.97 vs proposed (Real-ESRNet 27.75 vs 28.85 for MINC) | Hurt |
| dropout added to Real-ESRNet for generalization | Real-ESRNet · Synthetic setting ③ | PSNR: 26.88 vs 27.44 for Real-ESRNet | Hurt |
| feature extractor choice for Fréchet distance: DASR representation instead of AlexNet-MINC (ablation) | degradation weight estimation · Synthetic setting ① | PSNR: 28.88 vs 28.85 | No clear effect |
| random-init AlexNet as feature extractor instead of ImageNet/MINC-trained AlexNet (ablation) | degradation weight estimation · Synthetic setting ① | PSNR: 28.77 vs 28.85 (AlexNet-MINC) | Hurt |
| smaller alpha (wider training degradation range) vs larger alpha (narrower range) (ablation) | degradation distribution crafting · Synthetic test interval | PSNR within target range | Mixed |
Closest papers by meaning
- Not All Degradations Are Equal: A Targeted Feature Denoising Framework for Generalizable Image Super-Resolution (ICCV 2025 · similarity 0.90)
- Towards Realistic Data Generation for Real-World Super-Resolution (ICLR 2025 · similarity 0.90)
- SEAL: A Framework for Systematic Evaluation of Real-World Super-Resolution (ICLR 2024 · similarity 0.89)
- Unsupervised Diffusion-Based Degradation Modeling for Real-World Super-Resolution (AAAI 2025 · similarity 0.89)
- Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning (CVPR 2024 · similarity 0.88)
- Navigating Beyond Dropout: An Intriguing Solution towards Generalizable Image Super Resolution (CVPR 2024 · 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.
