Research map / Image generation / Image restoration
Learning to Push the Limits of Efficient FFT-Based Image Deconvolution
Jakob Kruse, Carsten Rother, Uwe Schmidt
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 |
|---|---|---|---|
| jointly finetune all stages end-to-end after greedy training | FDN10 · Levin et al. benchmark (sigma 1.5) | PSNR (dB): 34.98 -> 35.09 | Improved |
| jointly finetune all stages end-to-end after greedy training | FDN10 · Sun et al. benchmark (sigma 2.55) | PSNR (dB): 32.62 -> 32.67 | Improved |
| use proposed boundary adjustment instead of CSF-style edgetapering after each stage (ablation) | FDN10 G · Levin et al. benchmark | PSNR (dB): 0.74 dB | Improved |
| apply proposed boundary adjustment iteratively to Wiener filter instead of edgetapering (ablation) | Wiener filter · Levin et al. benchmark | PSNR (dB): over 1 dB | Improved |
| use CNN regularization (FDN) instead of CSF with edgetapering each stage, after 5 stages | FDN5 vs CSF5 · Levin et al. benchmark | PSNR (dB): 0.16 dB | Improved |
| use CNN regularization (FDN) instead of CSF with proposed boundary adjustment, after 5 stages | FDN5 vs CSF5 · Levin et al. benchmark | PSNR (dB): 0.32 dB | Improved |
| add stages 4-10 with proposed boundary adjustment (ablation) | FDN G · Levin et al. benchmark | PSNR (dB): 0.58 dB | Improved |
| add stages beyond 3 with edgetapering each stage (ablation) | FDN · Levin et al. benchmark | PSNR (dB) | No clear effect |
| use FDN10 T instead of EPLL/RTF | FDN10 T · Levin and Sun benchmarks | PSNR (dB): around 0.2 - 0.3 dB | Improved |
| use FDN instead of CSF | FDN10 · Sun and Levin benchmarks | PSNR (dB): around 0.5 and 1 dB | Improved |
Closest papers by meaning
- Learning Fully Convolutional Networks for Iterative Non-Blind Deconvolution (CVPR 2017 · similarity 0.89)
- Learning Discriminative Shrinkage Deep Networks for Image Deconvolution (ECCV 2022 · similarity 0.89)
- Fast Fourier Convolution (NeurIPS 2020 · similarity 0.88)
- Blind Image Deconvolution by Automatic Gradient Activation (CVPR 2016 · similarity 0.87)
- DWDN: Deep Wiener Deconvolution Network for Non-Blind Image Deblurring (TPAMI 2021 · similarity 0.87)
- Beyond Spatial Domain: Cross-domain Promoted Fourier Convolution Helps Single Image Dehazing (AAAI 2025 · similarity 0.85)
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.
