Research map / Image generation / Image restoration
Non-Blind Deblurring: Handling Kernel Uncertainty With CNNs
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 three estimates at different prior strengths instead of one estimate (ablation) | CNN2/3 · Dataset [22], kernels from [10] | PSNR: 29.50 -> 30.40 | Helped |
| Use three estimates at different prior strengths instead of one estimate (ablation) | CNN2/3 · Dataset [22], kernels from [5] | PSNR: 29.64 -> 30.62 | Helped |
| Use three estimates at different prior strengths instead of one estimate (ablation) | CNN2/3 · Dataset [22], kernels from [22] | PSNR: 29.91 -> 30.87 | Helped |
| Use hyper-Laplacian prior inputs (alpha=2/3) rather than Gaussian prior inputs (alpha=2) (ablation) | Three-input CNN deblurring · Dataset [22], kernels from [10] | PSNR: 29.91 -> 30.40 | Helped |
| Train with synthetic noisy kernels instead of real noisy kernels (ablation) | CNN deblurring · Test data | PSNR gain: 0.89 and 0.91 | No clear effect |
| Combine synthetic and real noisy kernels in training (ablation) | CNN deblurring · Test data | PSNR gain: 1.15 | Helped |
| Use the CNN to refine kernels returned by [5] (ablation) | Kernel estimation with [5] · Dataset [22], kernels from [5]+refinement | PSNR: 28.92 -> 31.19 | Helped |
| Use CNN restoration instead of the initial latent-image estimate for [40] (ablation) | [40] blind deblurring · Dataset [19], kernels from [40] | IFC: 2.00 -> 2.81 | Helped |
| Use three CNN input estimates instead of one (ablation) | CNN2/3 · Dataset [22] | PSNR | Helped |
| Add further prior-strength inputs beyond three (ablation) | CNN deblurring · Test data | PSNR | No clear effect |
| Train on only synthetic noisy kernels rather than combining kernel sources (ablation) | CNN deblurring · Test data | PSNR gain: 0.91 -> 1.15 | Hurt |
Closest papers by meaning
- Self-Supervised Non-Uniform Kernel Estimation With Flow-Based Motion Prior for Blind Image Deblurring (CVPR 2023 · similarity 0.94)
- Deep Learning for Handling Kernel/model Uncertainty in Image Deconvolution (CVPR 2020 · similarity 0.93)
- Parameterized Blur Kernel Prior Learning for Local Motion Deblurring (CVPR 2025 · similarity 0.93)
- Uncertainty-Aware Unsupervised Image Deblurring With Deep Residual Prior (CVPR 2023 · similarity 0.92)
- DeblurDiff: Real-Word Image Deblurring with Generative Diffusion Models (NeurIPS 2025 · similarity 0.91)
- Real-World Efficient Blind Motion Deblurring via Blur Pixel Discretization (CVPR 2024 · 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.