Research map / Image generation / Image restoration
Training deep learning based denoisers without ground truth data
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 |
|---|---|---|---|
| train with SURE loss instead of plain MSE to the noisy input (ablation) | SDA · MNIST sigma=50 | PSNR (dB): 19.85 -> 25.23 | Helped |
| use SURE training without ground truth instead of BM3D | SDA · MNIST sigma=50 | PSNR (dB): 21.82 -> 25.23 | Helped |
| add noisy test images to SURE training (SURE-T) (ablation) | SDA-SURE · MNIST sigma=25 | PSNR (dB): 28.35 -> 28.39 | No clear effect |
| SURE training (no ground truth) vs BM3D | DnCNN · BSD68 sigma=25 | PSNR (dB): 28.56 -> 28.97 | Helped |
| SURE training (no ground truth) vs BM3D | DnCNN · Set12 sigma=75 average | PSNR (dB): 24.89 -> 24.70 | Hurt |
| SURE training (no ground truth) vs BM3D | DnCNN · Set12 sigma=50 average | PSNR (dB): 26.67 -> 26.67 | No clear effect |
| add noisy test images to SURE training (SURE-T) (ablation) | DnCNN-SURE · BSD68 sigma=50 | PSNR (dB): 25.93 -> 25.95 | Helped |
| add noisy test images to SURE training (SURE-T) (ablation) | DnCNN-SURE · Set12 sigma=75 average | PSNR (dB): 24.70 -> 24.82 | Helped |
| train on BM3D outputs as pseudo ground truth instead of using BM3D directly | DnCNN · BSD68 sigma=50 | PSNR (dB): 25.62 -> 25.44 | Hurt |
| use noiseless ground truth MSE training instead of SURE | DnCNN · BSD68 sigma=50 | PSNR (dB): 25.93 -> 26.22 | Helped |
Closest papers by meaning
- Extending Stein's unbiased risk estimator to train deep denoisers with correlated pairs of noisy images (NeurIPS 2019 · similarity 0.94)
- High-Quality Self-Supervised Deep Image Denoising (NeurIPS 2019 · similarity 0.88)
- An Unsupervised Deep Learning Approach for Real-World Image Denoising (ICLR 2021 · similarity 0.88)
- UNSURE: self-supervised learning with Unknown Noise level and Stein's Unbiased Risk Estimate (ICLR 2025 · similarity 0.87)
- APR-RD: Complemental Two Steps for Self-Supervised Real Image Denoising (AAAI 2025 · similarity 0.86)
- Ditch the Denoiser: Emergence of Noise Robustness in Self-Supervised Learning from Data Curriculum (NeurIPS 2025 · similarity 0.85)
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.