Research map / Image generation / Super-resolution
KOALAnet: Blind Super-Resolution Using Kernel-Oriented Adaptive Local Adjustment
Soo Ye Kim, Hyeonjun Sim, Munchurl Kim
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 |
|---|---|---|---|
| use KOALAnet instead of the strongest listed baseline | Blind SR · random anisotropic degradation benchmarks, ×2 and ×4 | PSNR: 33.08 vs 28.61; 30.35 vs 26.63; 29.70 vs 26.86; 27.19 vs 24.11; 32.61 vs 26.19; 32.55 vs 28.90; 31.89 vs 30.36; 30.28 vs | Improved |
| add KOALA modules with k parameters to the upsampler (ablation) | KOALAnet · DIV2K-val ×4 | PSNR: 29.20 -> 29.40 | Improved |
| add m parameters to KOALA modules with k (ablation) | KOALAnet · DIV2K-val ×4 | PSNR: 29.40 -> 29.44 | Improved |
| replace estimated degradation kernels with ground-truth kernels (ablation) | KOALAnet · DIV2K-val ×4 | PSNR: 29.44 -> 29.67 | Improved |
| replace U-Net with ResNet in the downsampling network (ablation) | KOALAnet downsampler · DIV2K-val ×1/4 | PSNR: 49.50 -> 47.79 | Hurt |
| remove residual connections from downsampling-network ResBlocks (ablation) | KOALAnet downsampler · DIV2K-val ×1/4 | PSNR: 49.50 -> 47.85 | Hurt |
| estimate uniform instead of spatially-variant kernels (ablation) | KOALAnet downsampler · DIV2K-val ×1/4 | PSNR: 49.50 -> 45.78 | Hurt |
| remove normalization from predicted degradation kernels (ablation) | KOALAnet downsampler · DIV2K-val ×1/4 | PSNR: 49.50 -> 49.15 | Hurt |
| use the downsampling network's estimated spatially-variant kernels | KOALAnet · DIV2K-val | L2 kernel error: KernelGAN 0.0230; BlindSR 0.0152; KOALAnet 0.0010 | Improved |
| use kernel-aware local adjustment rather than an upsampler alone (ablation) | SR on shallow-DoF aesthetic images · artistic photographs with intentional blur | preservation of intended blur | Improved |
Later papers that compared with this method
1 later accepted paper reports KOALAnet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Uncertainty Learning in Kernel Estimation for Multi-stage Blind Image Super-Resolution (ECCV 2022 · better on 3 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
- End-to-End Alternating Optimization for Real-World Blind Super Resolution (IJCV 2023 · similarity 0.89)
- Blind Super-Resolution With Iterative Kernel Correction (CVPR 2019 · similarity 0.89)
- Blind Image Super-Resolution: A Survey and Beyond (TPAMI 2022 · similarity 0.89)
- KXNet: A Model-Driven Deep Neural Network for Blind Super-Resolution (ECCV 2022 · similarity 0.88)
- Unfolding the Alternating Optimization for Blind Super Resolution (NeurIPS 2020 · similarity 0.88)
- Spectrum-to-Kernel Translation for Accurate Blind Image Super-Resolution (NeurIPS 2021 · 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.
