Research map / Image generation / Image restoration
Blind2Sound: Self-Supervised Image Denoising without Residual Noise
Jiazheng Liu, Zejin Wang, Bohao Chen, Hua Han
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 |
|---|---|---|---|
| make masked and visible branches independent (IID) (ablation) | Blind2Sound · CBSD68 (alpha=0.01, sigma=0.0002) | PSNR (dB): 32.19 -> 33.14 | Improved |
| make masked and visible branches independent (IID) (ablation) | Blind2Sound · CBSD68 (alpha=0.05, sigma=0.02) | PSNR (dB): 28.95 -> 29.29 | Improved |
| use enhanced noise model M_E instead of M_O (ablation) | Blind2Sound · CBSD68 (alpha=0.05, sigma=0.02) | PSNR (dB): 29.12 -> 29.29 | Improved |
| use enhanced noise model M_E instead of M_S (ablation) | Blind2Sound · CBSD68 (alpha=0.05, sigma=0.02) | PSNR (dB): 29.14 -> 29.29 | Improved |
| joint training of noise estimator and denoiser instead of frozen pre-trained estimator (ablation) | Blind2Sound · CBSD68 (alpha=0.01, sigma=0.0002) | PSNR (dB): 32.78 -> 33.14 | Improved |
| joint training instead of fixed true noise (ablation) | Blind2Sound · CBSD68 (alpha=0.01, sigma=0.0002) | PSNR (dB): 33.02 -> 33.14 | Improved |
| joint training instead of frozen estimator (ablation) | Blind2Sound · CBSD68 (alpha=0.05, sigma=0.02) | PSNR (dB): 29.24 -> 29.29 | No clear effect |
| raise final visible factor from 3 to 11 (ablation) | Blind2Sound · CBSD68 (alpha=0.01, sigma=0.02) | PSNR (dB): 32.75 -> 32.89 | Improved |
| raise final visible factor from 11 to 40 (ablation) | Blind2Sound · CBSD68 (alpha=0.01, sigma=0.02) | PSNR (dB): 32.89 -> 32.76 | Hurt |
| add Cramer Gaussian loss regularization with weight 0.01 (ablation) | Blind2Sound · CBSD68 | PSNR (dB) | Improved |
| add fine-grained sub-block constraint to global Gaussian loss (ablation) | Cramer Gaussian loss · BSD68 noise estimation (alpha=0.05, sigma=0.02) | estimated (alpha, sigma): 0.039/0.057 -> 0.048/0.022 | Improved |
| use Blind2Sound instead of Blind2Unblind | Blind2Sound · SIDD RAW Benchmark | PSNR (dB): 50.79 -> 50.92 | Improved |
Closest papers by meaning
- Blind2Unblind: Self-Supervised Image Denoising With Visible Blind Spots (CVPR 2022 · similarity 0.92)
- Noise2Score: Tweedie’s Approach to Self-Supervised Image Denoising without Clean Images (NeurIPS 2021 · similarity 0.91)
- Convexity-Aware Noise Calibration: A Self-Supervised Framework for Noise-Level-Unknown Image Denoising (CVPR 2026 · similarity 0.90)
- FBI-Denoiser: Fast Blind Image Denoiser for Poisson-Gaussian Noise (CVPR 2021 · similarity 0.90)
- Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising (CVPR 2025 · similarity 0.89)
- APR-RD: Complemental Two Steps for Self-Supervised Real Image Denoising (AAAI 2025 · similarity 0.89)
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.
