Research map / Image generation / Super-resolution
Kernel Modeling Super-Resolution on Real Low-Resolution Images
Ruofan Zhou, Sabine Susstrunk
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 realistic blur-kernels estimated from real photographs to training (ablation) | KMSR · DIV2K val, Gaussian g1.6 x2 | PSNR (dB): 25.62 -> 27.63 | Improved |
| add realistic blur-kernels estimated from real photographs to training (ablation) | KMSR · DIV2K val, realistic kernels x2 | PSNR (dB): 25.29 -> 27.52 | Improved |
| add realistic blur-kernels estimated from real photographs to training (ablation) | KMSR · DIV2K val, Gaussian g1.25 x2 | PSNR (dB): 26.47 -> 27.94 | Improved |
| add GAN-generated kernels to the estimated kernel pool (ablation) | KMSR · DIV2K val, Gaussian g1.6 x2 | PSNR (dB): 27.02 -> 27.63 | Improved |
| add GAN-generated kernels to the estimated kernel pool (ablation) | KMSR · DIV2K val, Gaussian g1.25 x2 | PSNR (dB): 27.42 -> 27.94 | Improved |
| add GAN-generated kernels to the estimated kernel pool (ablation) | KMSR · DIV2K val, realistic kernels x2 | PSNR (dB): 27.10 -> 27.52 | Improved |
| add GAN-generated kernels to the estimated kernel pool (ablation) | KMSR · DIV2K val, Gaussian g1.7 x2 | PSNR (dB): 27.90 -> 27.15 | Hurt |
| add GAN-generated kernels to the estimated kernel pool (ablation) | KMSR · DIV2K val, bicubic kernel x2 | PSNR (dB): 33.28 -> 33.52 | Hurt |
| add GAN kernel augmentation versus simple geometric augmentation, averaged | KMSR · DIV2K val x2, average over kernels | PSNR (dB): 0.5dB | Improved |
| train on realistic kernel pool instead of bicubic-only training | KMSR vs DBPN · DIV2K val, bicubic kernel x2 | PSNR (dB): 33.84 -> 33.52 | Hurt |
| train on realistic kernel pool instead of bicubic-only training | KMSR vs DBPN · DIV2K val, Gaussian g1.6 x2 | PSNR (dB): 25.70 -> 27.63 | Improved |
| train on realistic kernel pool instead of bicubic-only training | KMSR vs DBPN · x2 zoom-in dataset | PSNR (dB): 27.42 -> 29.13 | Improved |
Closest papers by meaning
- Toward Real-World Single Image Super-Resolution: A New Benchmark and a New Model (ICCV 2019 · similarity 0.92)
- Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network (CVPR 2016 · similarity 0.92)
- Image Super-Resolution Using Deep Convolutional Networks (TPAMI 2016 · similarity 0.91)
- Unsupervised Real-World Super-Resolution: A Domain Adaptation Perspective (ICCV 2021 · similarity 0.91)
- Learning a Single Convolutional Super-Resolution Network for Multiple Degradations (CVPR 2018 · similarity 0.91)
- Correction Filter for Single Image Super-Resolution: Robustifying Off-the-Shelf Deep Super-Resolvers (CVPR 2020 · similarity 0.91)
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.
