Research map / Label-efficient and robust learning / Noisy labels
Noise2Self: Blind Denoising by Self-Supervision
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 UNet with self-supervised J-invariant loss instead of clean targets | UNet · Hànzì | PSNR: 13.1 $\pm$ 0.7 -> 13.8 $\pm$ 0.3 | Helped |
| Train DnCNN with self-supervised J-invariant loss instead of clean targets | DnCNN · ImageNet | PSNR: 22.0 -> 18.7 | Hurt |
| Train UNet with self-supervision instead of independently noisy targets | UNet · ImageNet | PSNR: 17.8 -> 18.6 | Helped |
| Train DnCNN with self-supervision instead of independently noisy targets | DnCNN · CellNet | PSNR: 34.4 $\pm$ 0.1 -> 33.7 $\pm$ 0.2 | Hurt |
| Tune a J-invariant median filter using self-supervised loss rather than its default parameter | Median · Single image | PSNR: 27.1 -> 27.5 | Helped |
| Tune a J-invariant wavelet filter using self-supervised loss rather than its default parameter | Wavelet · Single image | PSNR: 24.6 -> 26.0 | Helped |
| Tune J-invariant NL-Means using self-supervised loss rather than its default parameter | NL-Means · Single image | PSNR: 28.9 -> 30.4 | Helped |
| Mix the optimal amount of noisy input into the J-invariant output | J-invariant classical denoisers · Single image | PSNR: 27.5 -> 28.2; 26.0 -> 26.9; 30.4 -> 30.8 | Helped |
| Apply the trained CNN directly to noisy input instead of using its J-invariant version (ablation) | Noise2Self CNN · Deep-learning denoising datasets | PSNR: 0.5 dB | Helped |
| Use the self-supervised J-invariant predictor instead of the optimal unrestricted predictor | Optimal denoiser · Gaussian-process signals | reconstruction error | Mixed |
Closest papers by meaning
- Noise2Same: Optimizing A Self-Supervised Bound for Image Denoising (NeurIPS 2020 · similarity 0.90)
- Convexity-Aware Noise Calibration: A Self-Supervised Framework for Noise-Level-Unknown Image Denoising (CVPR 2026 · similarity 0.90)
- Noise2Score: Tweedie’s Approach to Self-Supervised Image Denoising without Clean Images (NeurIPS 2021 · similarity 0.90)
- Ditch the Denoiser: Emergence of Noise Robustness in Self-Supervised Learning from Data Curriculum (NeurIPS 2025 · similarity 0.89)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.88)
- Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising (CVPR 2025 · similarity 0.88)
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.