Research map / Image generation / Super-resolution
SRFlow: Learning the Super-Resolution Space with Normalizing Flow
Andreas Lugmayr, Martin Danelljan, Luc Van Gool, Radu Timofte
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 |
|---|---|---|---|
| pretrain the LR-Encoder (ablation) | SRFlow · CelebA 8x | PSNR: 25.06 -> 25.24 | Improved |
| pretrain the LR-Encoder (ablation) | SRFlow · CelebA 8x | LPIPS: 0.108 -> 0.110 | No clear effect |
| use SRFlow instead of ESRGAN | SRFlow · CelebA 8x (bicubic) | PSNR: 22.88 -> 25.24 | Improved |
| use SRFlow instead of ESRGAN | SRFlow · DIV2K 4x | PSNR: 26.22 -> 27.09 | Improved |
| use SRFlow instead of ESRGAN | SRFlow · DIV2K 8x | PSNR: 22.18 -> 23.05 | Improved |
| lower sampling temperature to tau=0 (ablation) | SRFlow · DIV2K / CelebA | PSNR: similar PSNR to RRDB for tau=0 | Mixed |
| raise sampling temperature to tau >= 0.8 (ablation) | SRFlow · perception-distortion | LPIPS: similar or better perceptual quality than ESRGAN for tau >= 0.8 | Improved |
| select best of n=1..10 samples (oracle) (ablation) | SRFlow tau=0.8 · CelebA test 5000 images | LPIPS | Improved |
| apply latent space normalization for restoration | SRFlow · DIV2K denoising sigma=20 | PSNR: 22.48 -> 27.81 | Improved |
| apply latent space normalization for restoration | SRFlow · CelebA denoising sigma=20 | PSNR: 22.52 -> 27.62 | Improved |
| super-resolve noisy image without normalization | SRFlow · DIV2K denoising sigma=20 | PSNR: 22.48 -> 23.19 | Improved |
Later papers that compared with this method
3 later accepted papers report SRFlow as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · better on 5 of 5 shared results)
- LAR-SR: A Local Autoregressive Model for Image Super-Resolution (CVPR 2022 · better on 1 of 1 shared result)
- Fourier Space Losses for Efficient Perceptual Image Super-Resolution (ICCV 2021 · better on 2 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
- Hierarchical Conditional Flow: A Unified Framework for Image Super-Resolution and Image Rescaling (ICCV 2021 · similarity 0.91)
- Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network (CVPR 2017 · similarity 0.91)
- Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · similarity 0.91)
- Deep Learning for Image Super-Resolution: A Survey (TPAMI 2020 · similarity 0.89)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.89)
- GSDD: Generative Space Dataset Distillation for Image Super-resolution (AAAI 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.
