Research map / Image generation / Super-resolution
Learning Controllable Degradation for Real-World Super-Resolution via Constrained Flows
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train SR networks on InterFlow-generated LR images for unseen levels (2<s<4) instead of only RealSR ×2,×4 | HAN, RCAN, VDSR, KPN, NLSN, SwinIR · RealSR ×3 test set | PSNR: e.g., HAN 30.43 -> 30.68; RCAN 30.30 -> 30.57 | Helped |
| Remove both LR-consistency and IB losses from InterFlow (ablation) | InterFlow training · RealSR ×3 test set (HAN) | PSNR: 30.68 -> 30.34 | Hurt |
| Add IB loss on top of LR-consistency loss (ablation) | InterFlow · RealSR ×3 test set (HAN) | PSNR: 30.16 -> 30.68 | Helped |
| Add LR-consistency loss on top of IB loss (ablation) | InterFlow · RealSR ×3 test set (HAN) | PSNR: 30.40 -> 30.68 | Helped |
| Replace latent-space interpolation with naive linear image-space interpolation of LR images (ablation) | HAN training data generation · RealSR ×3 test set | PSNR: 30.68 (InterFlow) vs 30.48 (linear) | Hurt |
| Use InterFlow-generated LR data to train scale-arbitrary networks (MetaSR, LIIF) instead of bicubic or Real-ESRGAN/BSRGAN | MetaSR, LIIF · RealSR ×3 test set | PSNR: MetaSR 28.99 (bicubic) -> 30.42; LIIF 29.00 -> 30.44 | Helped |
| Use InterFlow data instead of BSRGAN/Real-ESRGAN data for MetaSR and LIIF | MetaSR, LIIF · DRealSR ×3 test set | PSNR: MetaSR 30.07 (BSRGAN) -> 31.16; LIIF 30.21 -> 31.15 | Helped |
| Train on InterFlow data for RealSR ×2.5 and ×3.5 test sets | RCAN, NLSN · RealSR ×2.5 / ×3.5 test sets | PSNR: RCAN ×2.5 31.79 -> 32.11; ×3.5 30.06 -> 30.20 | Helped |
| Increase LR-consistency loss weight from 10 to 100 (ablation) | InterFlow · RealSR ×3 test set (HAN) | PSNR: 30.68 -> 30.52 | Hurt |
| Use 2x downsampling instead of 4x in LR-consistency loss (ablation) | InterFlow · RealSR ×3 test set (HAN) | PSNR: 30.68 -> 30.25 | Hurt |
Closest papers by meaning
- 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)
- Towards Realistic Data Generation for Real-World Super-Resolution (ICLR 2025 · similarity 0.91)
- Unsupervised Degradation Representation Aware Transform for Real-World Blind Image Super-Resolution (AAAI 2025 · similarity 0.90)
- Deep Generative Model based Rate-Distortion for Image Downscaling Assessment (CVPR 2024 · similarity 0.89)
- DORNet: A Degradation Oriented and Regularized Network for Blind Depth Super-Resolution (CVPR 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.