Research map / Image generation / Image restoration
Extending Stein's unbiased risk estimator to train deep denoisers with correlated pairs of noisy images
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 eSURE instead of SURE* with two independent noisy realizations | DnCNN · BSD-68 and Set 12, σ=25 and σ=50 | PSNR (dB): 28.92 -> 29.08; 26.00 -> 26.15; 30.04 -> 30.31; 26.87 -> 27.07 | Helped |
| Use eSURE instead of Noise2Noise with independent noisy pairs | DnCNN · BSD-68 and Set 12, σ=25 and σ=50 | PSNR (dB): 29.08 vs 29.08; 26.15 vs 26.13; 30.31 vs 30.30; 27.07 vs 27.07 | No clear effect |
| Train SURE with twice as many noisy examples (SURE*) | DnCNN-SURE · BSD-68 and Set 12, σ=25 and σ=50 | PSNR (dB): 28.92 -> 29.00; 26.00 -> 26.07; 30.04 -> 30.13; 26.87 -> 26.97 | Helped |
| Use eSURE rather than Noise2Noise with more correlated training pairs (σ_gt=10) | DnCNN · BSD-68, σ_noisy=25 and 50 | PSNR (dB): 28.37 -> 29.21; 25.91 -> 26.27 | Helped |
| Use eSURE rather than Noise2Noise with more correlated training pairs (σ_gt=10) | Color DnCNN · RGB BSD-68, σ_noisy=25 and 50 | PSNR (dB): 29.83 -> 31.19; 27.61 -> 27.90 | Helped |
| Increase ground-truth noise from σ_gt=1 to σ_gt=10 when training with Noise2Noise | DnCNN-N2N · BSD-68, σ_noisy=25 and 50 | PSNR (dB): 29.23 -> 28.37; 26.28 -> 25.91 | Hurt |
| Increase ground-truth noise from σ_gt=1 to σ_gt=10 when training with Noise2Noise | CDnCNN-N2N · RGB BSD-68, σ_noisy=25 and 50 | PSNR (dB): 31.18 -> 29.83; 27.89 -> 27.61 | Hurt |
Closest papers by meaning
- Training deep learning based denoisers without ground truth data (NeurIPS 2018 · similarity 0.94)
- Noisier2Noise: Learning to Denoise From Unpaired Noisy Data (CVPR 2020 · similarity 0.89)
- UNSURE: self-supervised learning with Unknown Noise level and Stein's Unbiased Risk Estimate (ICLR 2025 · similarity 0.89)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.87)
- Ditch the Denoiser: Emergence of Noise Robustness in Self-Supervised Learning from Data Curriculum (NeurIPS 2025 · similarity 0.86)
- On the Posterior Distribution in Denoising: Application to Uncertainty Quantification (ICLR 2024 · 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.