Research map / Image generation / Image restoration
Learning to Deblur Images with Exemplars
Jinshan Pan, Wenqi Ren, Zhe Hu, Ming–Hsuan Yang
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 |
|---|---|---|---|
| replace exemplar search with CNN edge prediction (prediction step) | edge prediction · 480 frontal-face test images | prediction run time (s): 4260 -> 0.95 | Improved |
| replace exemplar search with CNN edge prediction (total) | deblurring pipeline · 480 frontal-face test images | total run time (s): 4287 -> 27.95 | Improved |
| use CNN-predicted edges instead of exemplar edges | kernel estimation · frontal face synthetic set | cumulative error ratio | Mixed |
| add predicted salient edges to kernel estimation (ablation) | kernel estimation · frontal face synthetic set | cumulative error ratio | Improved |
| add coarse-to-fine strategy (ablation) | kernel estimation without predicted edges · frontal face synthetic set | cumulative error ratio | No clear effect |
| use blurred-input edges instead of matched exemplar edges (ablation) | kernel estimation · frontal face synthetic set | cumulative error ratio | No clear effect |
| use 3x3 filters in all layers with same receptive field (ablation) | CNN edge predictor · frontal face dataset | avg PSNR: 35.33 -> 35.55 | No clear effect |
| vary first-layer filter size from 9x9 to 21x21 (ablation) | CNN edge predictor · blur kernel sizes 9x9 to 27x27 | PSNR | No clear effect |
| increase exemplar set size from 40 to 200 centers (ablation) | exemplar-based method · 40 blurred test images | cumulative error ratio | No clear effect |
| increase CNN training images from 200 to 1500 (ablation) | CNN-based method · 40 blurred test images | cumulative error ratio | No clear effect |
| iteratively update exemplar edges (ablation) | exemplar-based method · face deblurring | kernel estimation quality | Improved |
| add 1% or other noise to blurred inputs (ablation) | matching criterion · exemplar matching | matching accuracy | No clear effect |
Closest papers by meaning
- Exploiting Semantics for Face Image Deblurring (IJCV 2020 · similarity 0.92)
- Deep Semantic Face Deblurring (CVPR 2018 · similarity 0.91)
- Deep Image Deblurring: A Survey (IJCV 2022 · similarity 0.91)
- DeblurDiff: Real-Word Image Deblurring with Generative Diffusion Models (NeurIPS 2025 · similarity 0.86)
- Prior and Prediction Inverse Kernel Transformer for Single Image Defocus Deblurring (AAAI 2024 · similarity 0.86)
- Residual Diffusion Deblurring Model for Single Image Defocus Deblurring (AAAI 2025 · similarity 0.86)
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.
