Research map / Image generation / Image restoration
Learning Deep Non-Blind Image Deconvolution without Ground Truths
Yuhui Quan, Zhuojie Chen, Huan Zheng, Hui Ji
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 |
|---|---|---|---|
| add loss Lr to training (vs Lp only) (ablation) | UNID · Motion deblurring, Sun et al. kernel [8] | PSNR: 28.04 -> 30.24 | Improved |
| add loss Lp to training (vs Lr only) (ablation) | UNID · Motion deblurring, Sun et al. kernel [8] | PSNR: 29.92 -> 30.24 | Improved |
| add loss Lp to training (vs Lr only) (ablation) | UNID · Motion deblurring, Levin et al. kernel [31] | PSNR: 30.96 -> 31.71 | Improved |
| change Lr to Eq. 12 during training (ablation) | UNID · Motion deblurring, Sun et al. kernel [8] | PSNR: 30.13 -> 30.24 | Improved |
| train with GT images (supervised Lgt) instead of UNID (ablation) | UNID · Motion deblurring, Sun et al. kernel [8] | PSNR: 30.24 -> 30.27 | No clear effect |
| add proposed kernel perturbation at inference (ensemble) (ablation) | UNID · Motion deblurring, Sun et al. kernel [8] | PSNR: 29.31 -> 30.24 | Improved |
| add proposed kernel perturbation at inference (ensemble) (ablation) | UNID · Motion deblurring, Lai et al. kernel [48] | PSNR: 21.54 -> 22.43 | Improved |
| use proposed perturbation instead of Gaussian noise perturbation (ablation) | UNID · Motion deblurring, Sun et al. kernel [8] | PSNR: 29.48 -> 30.24 | Improved |
| use UNID instead of SURE loss | NN trained without GT · Motion deblurring, Sun et al. kernel [8] | PSNR: 26.91 -> 30.24 | Improved |
| use UNID instead of EI loss | NN trained without GT · Microscopic deconvolution, sigma 1% | PSNR: 36.87 -> 38.36 | Improved |
| train UNID without GT instead of supervised WF | UNID · Microscopic deconvolution, sigma 1% | PSNR: 37.81 -> 38.36 | Improved |
| use UNID instead of DWDN with accurate kernels | UNID · Motion deblurring, accurate kernels, Levin et al. 3% noise | PSNR: 29.85 -> 29.71 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- DWDN — DWDN: Deep Wiener Deconvolution Network for Non-Blind Image Deblurring (TPAMI 2021) · better on 1 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
- Learning Fully Convolutional Networks for Iterative Non-Blind Deconvolution (CVPR 2017 · similarity 0.90)
- Deep Learning for Handling Kernel/model Uncertainty in Image Deconvolution (CVPR 2020 · similarity 0.88)
- Training Image Estimators without Image Ground Truth (NeurIPS 2019 · similarity 0.88)
- Neural Blind Deconvolution Using Deep Priors (CVPR 2020 · similarity 0.88)
- Cross-Scale Self-Supervised Blind Image Deblurring via Implicit Neural Representation (NeurIPS 2024 · similarity 0.86)
- Unsupervised Blind Image Deblurring Based on Self-Enhancement (CVPR 2024 · 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.
