Research map / Image generation / Image restoration
Fully Trainable and Interpretable Non-Local Sparse Models for Image Restoration
Bruno Lecouat, Jean Ponce, Julien Mairal
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 non-local group sparsity (GroupSC) to the sparse coding baseline (ablation) | SC · BSD68 grayscale denoising (sigma=25) | PSNR (dB): 28.90 -> 29.20 | Improved |
| add non-local group sparsity (GroupSC) to the sparse coding baseline (ablation) | SC · BSD68 grayscale denoising (sigma=50) | PSNR (dB): 25.84 -> 26.17 | Improved |
| add non-local group sparsity (GroupSC) to the sparse coding baseline (ablation) | SC · CBSD68 color denoising (sigma=50) | PSNR (dB): 27.39 -> 28.05 | Improved |
| add non-local group sparsity (GroupSC) to the sparse coding baseline (ablation) | SC · Classic5 JPEG deblocking (qf=10) | PSNR (dB): 29.39 -> 29.61 | Improved |
| add non-local group sparsity (GroupSC) to the sparse coding baseline (ablation) | SC · Demosaicking Urban100 | PSNR (dB): 37.50 -> 38.21 | Improved |
| add non-local group sparsity (GroupSC) to the sparse coding baseline (ablation) | SC · CBSD68 blind color denoising (sigma=25) | PSNR (dB): 30.91 -> 31.25 | Improved |
| add similarity refinement updates during iterations (ablation) | GroupSC · BSD68 grayscale denoising | PSNR (dB): 0.15 dB | Improved |
| use a larger dictionary (128 to 512 atoms) (ablation) | GroupSC light · BSD68 grayscale denoising (sigma=25, patch k=9) | PSNR (dB): 29.12 -> 29.20 | Improved |
| increase patch size from 7 to 11 (n=256) (ablation) | GroupSC light · BSD68 grayscale denoising (sigma=5 and 25) | PSNR (dB): 37.92 -> 37.89 (sigma=5); 29.11 -> 29.18 (sigma=25) | Mixed |
| increase number of unrolled iterations from 12 to 36 (ablation) | SC / GroupSC light · BSD68 grayscale denoising | PSNR (dB): 0.02dB | No clear effect |
| reduce stride between image blocks from 56 to 12 (ablation) | GroupSC (6 updates, middle averaging) · BSD68 grayscale denoising (sigma=15) | PSNR (dB): 31.39 -> 31.71 | Improved |
| add middle averaging of similarity (stride 56) (ablation) | GroupSC · BSD68 grayscale denoising (sigma=15) | PSNR (dB): 31.38 -> 31.39 | No clear effect |
Closest papers by meaning
- Unsupervised Image Restoration Using Partially Linear Denoisers (TPAMI 2021 · similarity 0.92)
- Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration (ICLR 2025 · similarity 0.89)
- Stochastic Deep Restoration Priors for Imaging Inverse Problems (ICML 2025 · similarity 0.88)
- SeNM-VAE: Semi-Supervised Noise Modeling with Hierarchical Variational Autoencoder (CVPR 2024 · similarity 0.88)
- A Restoration Network as an Implicit Prior (ICLR 2024 · similarity 0.87)
- Adapt or Perish: Adaptive Sparse Transformer with Attentive Feature Refinement for Image Restoration (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.
