Research map / Image generation / Super-resolution
Continuous Degradation Modeling via Latent Flow Matching for Real-World Super-Resolution
Hyeonjae Kim, Dongjin Kim, Eugene Jin, Tae Hyun Kim
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 |
|---|---|---|---|
| replace piecewise linear latent trajectory with natural cubic spline (ablation) | DegFlow (HAT SR) · RealSR x3 test | PSNR / LPIPS: PSNR 30.58 -> 30.68; LPIPS 0.3214 -> 0.3209 | Improved |
| add LPIPS loss with 3rd-order Taylor extrapolation (ablation) | DegFlow (HAT SR) · RealSR x3 test | PSNR / LPIPS: PSNR 30.68 -> 30.81; LPIPS 0.3209 -> 0.3200 | Improved |
| add HR-feature skip connections in RAE (ablation) | DegFlow (HAT SR) · RealSR x3 test | PSNR / LPIPS: PSNR 30.81 -> 30.86; LPIPS 0.3200 -> 0.3186 | Improved |
| add DIV2K synthetic LR data generated by DegFlow (ablation) | HAT · RealSR x3 test | PSNR: 30.86 -> 31.00 | Improved |
| train with DegFlow data instead of InterFlow data | MambaIR · RealSR x3 test | PSNR: 30.51 -> 30.73 | Improved |
| train with DegFlow x2~x4 data instead of real x2,x4 pairs | HAT · RealSR x3 test | PSNR: 30.39 -> 30.86 | Improved |
| DegFlow vs InterFlow data on LPIPS (fixed-scale) | HAT · RealSR x3 test | LPIPS: 0.3135 -> 0.3186 | Hurt |
| DegFlow data instead of InterFlow for LIIF | LIIF · RealArbiSR x3 | PSNR: 29.38 -> 29.74 | Improved |
Closest papers by meaning
- Learning Controllable Degradation for Real-World Super-Resolution via Constrained Flows (ICML 2023 · similarity 0.95)
- Unsupervised Diffusion-Based Degradation Modeling for Real-World Super-Resolution (AAAI 2025 · similarity 0.92)
- Low-Res Leads the Way: Improving Generalization for Super-Resolution by Self-Supervised Learning (CVPR 2024 · similarity 0.91)
- Unsupervised Degradation Representation Aware Transform for Real-World Blind Image Super-Resolution (AAAI 2025 · similarity 0.90)
- Towards Realistic Data Generation for Real-World Super-Resolution (ICLR 2025 · similarity 0.89)
- Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · 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.
