Research map / Label-efficient and robust learning / Self-supervised and contrastive
Noise Distribution Adaptive Self-Supervised Image Denoising Using Tweedie Distribution and Score Matching
Check 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 |
|---|---|---|---|
| Add EMA to score-model training (ablation) | Proposed denoising method · CBSD68, Poisson ζ=0.01 | PSNR (dB): 32.23 -> 32.53 | Helped |
| Add geometric-sequence annealing to score-model training (ablation) | Proposed denoising method · CBSD68, Poisson ζ=0.01 | PSNR (dB): 32.41 -> 32.53 | Helped |
| Add both EMA and geometric-sequence annealing to score-model training (ablation) | Proposed denoising method · CBSD68, Poisson ζ=0.01 | PSNR (dB): 32.03 -> 32.53 | Helped |
| Use the proposed noise-level estimator instead of the quality-penalty estimator (ablation) | Proposed denoising method · CBSD68, Gaussian σ=25 | PSNR (dB): 30.78 -> 30.89 | Helped |
| Use the proposed noise-level estimator instead of the quality-penalty estimator (ablation) | Proposed denoising method · CBSD68, Poisson ζ=0.01 | PSNR (dB): 31.89 -> 32.53 | Helped |
| Use the proposed noise-level estimator instead of the quality-penalty estimator (ablation) | Proposed denoising method · CBSD68, Gamma k=100 | PSNR (dB): 33.92 -> 34.52 | Helped |
| Use the proposed blind noise-level estimator rather than known noise level (ablation) | Proposed denoising method · CBSD68, Gaussian σ=25 | PSNR (dB): 30.89 vs 30.91 | No clear effect |
| Use the proposed blind noise-level estimator rather than known noise level (ablation) | Proposed denoising method · CBSD68, Poisson ζ=0.01 | PSNR (dB): 32.53 vs 32.63 | No clear effect |
| Use the proposed blind noise-level estimator rather than known noise level (ablation) | Proposed denoising method · CBSD68, Gamma k=100 | PSNR (dB): 34.52 vs 34.53 | No clear effect |
| Estimate noise level with the proposed method instead of quality penalty (ablation) | Noise2Score-based denoising · CBSD68 inference | inference time: 3.1s -> 0.1s | Helped |
| Use the proposed blind method rather than the strongest self-supervised baseline | Image denoising · CBSD68 and Kodak, Gaussian σ=25 and 50 | PSNR (dB): 30.89 vs 30.92; 31.95 vs 31.96; 27.56 vs 27.65; 28.64 vs 28.63 | Mixed |
| Use the proposed blind method rather than the strongest self-supervised baseline | Image denoising · CBSD68 and Kodak, Poisson ζ=0.01 and 0.05 | PSNR (dB): 32.53 vs 32.73; 33.41 vs 33.51; 28.73 vs 28.83; 29.72 vs 30.17 | Hurt |
Closest papers by meaning
- On the Posterior Distribution in Denoising: Application to Uncertainty Quantification (ICLR 2024 · similarity 0.88)
- Energy-Tweedie: Score meets Score, Energy meets Energy (NeurIPS 2026 · similarity 0.87)
- Sampling Binary Data by Denoising through Score Functions (ICML 2025 · similarity 0.85)
- Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising (CVPR 2025 · similarity 0.85)
- APR-RD: Complemental Two Steps for Self-Supervised Real Image Denoising (AAAI 2025 · similarity 0.84)
- Exploring Efficient Asymmetric Blind-Spots for Self-Supervised Denoising in Real-World Scenarios (CVPR 2024 · similarity 0.84)
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.