Research map / Image generation / Image restoration
End-to-end Interpretable Learning of Non-blind Image Deblurring
Thomas Eboli, Jian Sun, Jean Ponce
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 |
|---|---|---|---|
| solve least squares with CPCR instead of conjugate gradient (ablation) | HQS TV-l1 · 160 synthetic images | average PSNR (dB): 26.12 -> 26.27 | Improved |
| use CPCR instead of FFT with FDN padding (ablation) | HQS TV-l1 · 160 synthetic images | average PSNR (dB): 25.96 -> 26.27 | Improved |
| use edge padding for FFT deconvolution (ablation) | HQS-FFT · 160 synthetic images | average PSNR (dB): 20.69 -> 25.53 | Improved |
| increase inverse filter size ratio w_c0/w_k up to 2.2 (ablation) | CPCR preconditioner · 160 images, lambda=0 | PSNR | Improved |
| learn the CNN prox and preconditioner modules (LCHQS_F vs CHQS) | CHQS · Levin | PSNR (dB): 32.12 -> 35.15 | Improved |
| learn the CNN prox and preconditioner modules (LCHQS_F vs CHQS) | CHQS · Sun | PSNR (dB): 30.36 -> 32.93 | Improved |
| add final-output-only fine-tuning stage (LCHQS_F vs LCHQS_G) | LCHQS_G · Levin / Sun | PSNR (dB): 35.11 -> 35.15; 32.83 -> 32.93 | Improved |
| add final-output-only fine-tuning stage (ablation) | LCHQS blind · uniform deblurring 5% noise | PSNR (dB): 24.05 -> 24.21 | Improved |
| add final-output-only fine-tuning stage (ablation) | LCHQS non-blind · uniform deblurring 5% noise | PSNR (dB): 23.87 -> 23.95 | No clear effect |
| train a single blind model over noise levels instead of noise-specific models (ablation) | LCHQS_G · uniform deblurring 1% / 5% noise | PSNR (dB): 29.54 -> 29.22; 23.87 -> 24.05 | Mixed |
| use 2/3 CPCR iterations instead of 1 (ablation) | LCHQS_G · Levin and Sun | PSNR (dB): +0.4/0.5dB over FDN | Improved |
| use LCHQS instead of CG-based method | non-uniform deblurring · 100 PASCAL images 1% noise | PSNR (dB): 25.84 -> 26.98 | Improved |
Closest papers by meaning
- Learning a Discriminative Prior for Blind Image Deblurring (CVPR 2018 · similarity 0.90)
- Non-Blind Deblurring: Handling Kernel Uncertainty With CNNs (CVPR 2018 · similarity 0.90)
- Uncertainty-Aware Unsupervised Image Deblurring With Deep Residual Prior (CVPR 2023 · similarity 0.89)
- Cross-Scale Self-Supervised Blind Image Deblurring via Implicit Neural Representation (NeurIPS 2024 · similarity 0.89)
- DeblurDiff: Real-Word Image Deblurring with Generative Diffusion Models (NeurIPS 2025 · similarity 0.88)
- Unsupervised Blind Image Deblurring Based on Self-Enhancement (CVPR 2024 · similarity 0.87)
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.
