Research map / Image generation / Image restoration
Unsupervised Image Restoration Using Partially Linear Denoisers
Rihuan Ke, Carola‐Bibiane Schönlieb
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 |
|---|---|---|---|
| train with DPLD instead of Noise2Self (sigma=25) | DnCNN network, noisy-only training · BSD68 Gaussian | PSNR (dB): 27.48 -> 29.08 | Improved |
| train with DPLD instead of Noise2Self (sigma=50) | DnCNN network, noisy-only training · BSD68 Gaussian | PSNR (dB): 25.15 -> 26.13 | Improved |
| train with DPLD instead of SURE | DnCNN network, noisy-only training · BSD68 Gaussian sigma=25 | PSNR (dB): 28.99 -> 29.08 | Improved |
| train with DPLD instead of Noisier2Noise | DnCNN network, noisy-only training · BSD68 Gaussian sigma=50 | PSNR (dB): 25.96 -> 26.13 | Improved |
| use ground truth supervised training instead of DPLD | DPLD · BSD68 Gaussian | PSNR (dB): 29.08 -> 29.22 | Improved |
| train with DPLD instead of Noise2Self (Poisson) | DnCNN network, noisy-only training · BSD68 Poisson | PSNR (dB): 27.78 -> 29.28 | Improved |
| use ground truth supervised training instead of DPLD (Poisson) | DPLD · BSD68 Poisson | PSNR (dB): 29.28 -> 29.43 | Improved |
| add partially linear constraint with very small gamma (1e-4) versus tuned gamma (ablation) | DPLD · BSD68 Poisson lambda=30 | PSNR (dB) | Improved |
| vary gamma between 2^2 and 2^8 (ablation) | DPLD · BSD68 Poisson lambda=30 | PSNR (dB): less than 0.2 dB | No clear effect |
| use very large gamma (2^14) (ablation) | DPLD · BSD68 Poisson lambda=30 | PSNR (dB) | Hurt |
| underestimate or overestimate the noise variance used to generate z (4% less) (ablation) | DPLD · BSD68 Gaussian | PSNR (dB): drop below 0.1 dB | Hurt |
| modify denoiser to g(x)+L n (remove residual e) on parrot (ablation) | DnCNN · Parrot Gaussian sigma=25 | PSNR (dB): 29.47 -> 29.60 | Improved |
Closest papers by meaning
- Universal Denoising Networks : A Novel CNN Architecture for Image Denoising (CVPR 2018 · similarity 0.93)
- IDR: Self-Supervised Image Denoising via Iterative Data Refinement (CVPR 2022 · similarity 0.92)
- Unpaired Learning of Deep Image Denoising (ECCV 2020 · similarity 0.92)
- High-Quality Self-Supervised Deep Image Denoising (NeurIPS 2019 · similarity 0.92)
- Denoising Prior Driven Deep Neural Network for Image Restoration (TPAMI 2018 · similarity 0.92)
- Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration (ICLR 2025 · similarity 0.90)
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.
