Research map / Image generation / Image restoration
Normalized Blind Deconvolution
Meiguang Jin, Stefan Roth, Paolo Favaro
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 exact L2 blur normalization instead of L1 normalization (ablation) | PN on BSDS · small BSDS dataset (72 blurry images) | mean SSD error ratio: 2.211 -> 2.011 | Improved |
| add exact L2 blur normalization instead of L1 normalization (ablation) | PN on BSDS · small BSDS dataset (72 blurry images) | failure cases: 3 -> 0 | Improved |
| use principled normalization instead of heuristic weight decay scaled by kernel norm (ablation) | PN on BSDS · small BSDS dataset (72 blurry images) | mean SSD error ratio: 2.591 -> 2.011 | Improved |
| use normalization instead of additive explicit L2 blur prior (ablation) | PN on BSDS · small BSDS dataset (72 blurry images) | mean SSD error ratio: 2.487 -> 2.011 | Improved |
| use p=2 norm instead of larger p in the normalized prior (ablation) | PN and FW · 12 BSDS images, 6 blurs | SSD error ratio: p = 2 performs best among the different Lp norms | Improved |
| use p=2 instead of p=1 in the normalized prior (FW method) (ablation) | FW · 12 BSDS images, 6 blurs | SSD error ratio: p=1 gets stuck at the trivial solution | Improved |
| increase p beyond 2 (ablation) | blur estimate · BSDS blur bias | weight alpha of true kernel: alpha decreases and kernel moves toward uniform | Hurt |
| use PN instead of Pan et al. on full BSDS | blind deblurring · full BSDS dataset (3000 blurry images) | mean SSD error ratio: 2.956 -> 2.067 | Improved |
| use PN instead of Pan et al. on full BSDS | blind deblurring · full BSDS dataset (3000 blurry images) | failure cases: 325 -> 94 | Improved |
| use PN instead of Pan et al. on SUN | blind deblurring · SUN dataset (640 blurry images) | mean SSD error ratio: 1.914 -> 2.299 | Hurt |
| use PN instead of Perrone & Favaro on SUN | blind deblurring · SUN dataset (640 blurry images) | maximum error ratio: 8.517 -> 6.764 | Improved |
| use PN instead of Pan et al. under 5% noise | blind deblurring · Levin dataset (32 images) | normalized L2 error: 5.368 -> 4.202 | Improved |
Closest papers by meaning
- Convergence Analysis of MAP Based Blur Kernel Estimation (ICCV 2017 · similarity 0.91)
- Kernel Diffusion: An Alternate Approach to Blind Deconvolution (ECCV 2024 · similarity 0.91)
- Blind Image Deconvolution by Automatic Gradient Activation (CVPR 2016 · similarity 0.90)
- A Clearer Picture of Total Variation Blind Deconvolution (TPAMI 2016 · similarity 0.90)
- Blind image deblurring with noise-robust kernel estimation (ECCV 2024 · similarity 0.90)
- Neural Blind Deconvolution Using Deep Priors (CVPR 2020 · similarity 0.89)
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.
