Research map / Image generation / Super-resolution
Learning Many-to-Many Mapping for Unpaired Real-World Image Super-Resolution and Downscaling
Wanjie Sun, Zhenzhong Chen
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 backward loss (perceptual and GAN) at tau=0 | SDFlow · RealSR 4x SR | LPIPS: 0.2934 -> 0.2500 | Improved |
| add backward loss (perceptual and GAN) at tau=0.8 | SDFlow · RealSR 4x SR | NIQE: 5.7145 -> 4.8106 | Improved |
| add backward loss at tau=0.8 | SDFlow · DRealSR 4x SR | NIQE: 7.6900 -> 5.5167 | Improved |
| raise sampling temperature from 0 to 0.8 (forward loss) | SDFlow · RealSR 4x SR | LPIPS: 0.2934 -> 0.2321 | Improved |
| raise sampling temperature from 0 to 0.8 (forward loss) | SDFlow · RealSR 4x SR | PSNR: 28.5813 -> 27.8403 | Hurt |
| raise sampling temperature from 0 to 0.8 (forward loss) | SDFlow · DRealSR 4x SR | LPIPS: 0.3411 -> 0.2852 | Improved |
| add backward loss at tau=0.8 | SDFlow · DRealSR 4x SR | PSNR: 30.5901 -> 29.4299 | Hurt |
| use SDFlow instead of ADL for downscaling | downscaling model · RealSR 4x downscaling | PSNR: 34.5064 -> 34.8498 | Improved |
| use SDFlow instead of DeFlow for synthesizing LR training pairs | ESRGAN · RealSR 4x SR (ESRGAN trained on synthesized pairs) | LPIPS: 0.3356 -> 0.2262 | Improved |
| use SDFlow instead of DeFlow for synthesizing LR training pairs | ESRGAN · RealSR 4x SR (ESRGAN trained on synthesized pairs) | NIQE: 7.2015 -> 5.2521 | Improved |
| SDFlow vs DeFlow downscaling diversity | downscaling model · RealSR 4x downscaling | Diversity: 10.1595 -> 2.2314 | Hurt |
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.
- DeFlow — DeFlow: Learning Complex Image Degradations From Unpaired Data With Conditional Flows (CVPR 2021) · better on 1 of 3 shared results
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
- Learning Controllable Degradation for Real-World Super-Resolution via Constrained Flows (ICML 2023 · similarity 0.94)
- Unsupervised Diffusion-Based Degradation Modeling for Real-World Super-Resolution (AAAI 2025 · similarity 0.91)
- Towards Realistic Data Generation for Real-World Super-Resolution (ICLR 2025 · similarity 0.91)
- Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning (CVPR 2024 · similarity 0.90)
- Learning Coupled Dictionaries from Unpaired Data for Image Super-Resolution (CVPR 2024 · similarity 0.89)
- Unsupervised Degradation Representation Aware Transform for Real-World Blind Image Super-Resolution (AAAI 2025 · similarity 0.89)
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.
