Research map / Image generation / Fast diffusion sampling
Unsupervised Image Denoising with Score Function
Yutong Xie, Mingze Yuan, Bin Dong, Quanzheng Li
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 |
|---|---|---|---|
| Use score-function method instead of Neighbor2Neighbor with correlation transform A | Gaussian sigma=25 w/ A · Kodak | PSNR (dB): 27.89 -> 33.99 | Improved |
| Use score-function method instead of Noisier2Noise | Gaussian sigma=25 w/o A · Kodak | PSNR (dB): 31.80 -> 31.92 | No clear effect |
| Use score-function method instead of Neighbor2Neighbor on signal-dependent Gaussian noise w/o A | Gaussian a=0.98 b=25 · Kodak | PSNR (dB): 30.42 -> 29.68 | Hurt |
| Use score-function method instead of Neighbor2Neighbor on signal-dependent Gaussian noise w/ A | Gaussian a=0.98 b=25 · Kodak | PSNR (dB): 25.02 -> 31.88 | Improved |
| Use score-function method instead of Neighbor2Neighbor on unbiased Gamma noise | Gamma a=26 w/o A · Kodak | PSNR (dB): 32.98 -> 32.61 | Hurt |
| Use score-function method instead of Neighbor2Neighbor on unbiased Poisson noise | Poisson l=0.2 w/o A · Kodak | PSNR (dB): 32.50 -> 32.38 | Hurt |
| Use score-function method instead of Neighbor2Neighbor on biased Rayleigh noise | Rayleigh s=0.3 w/o A · Kodak | PSNR (dB): 16.55 -> 34.25 | Improved |
| Use score-function method instead of Neighbor2Neighbor on mixture Gamma noise w/ A | Gamma a=26 + Gaussian, w/ A · Kodak | PSNR (dB): 26.93 -> 30.74 | Improved |
| Use score-function method instead of Neighbor2Neighbor on mixture Poisson noise w/o A | Poisson l=0.2 + Gaussian, w/o A · Kodak | PSNR (dB): 32.03 -> 32.10 | No clear effect |
| Use score-function method instead of Neighbor2Neighbor on mixture Gamma noise w/o A | Gamma a=26 + Gaussian, w/o A · Kodak | PSNR (dB): 32.30 -> 32.13 | Hurt |
| Use score-function method instead of Neighbor2Neighbor on mixture Rayleigh noise w/ A | Rayleigh s=0.3 + Gaussian, w/ A · Kodak | PSNR (dB): 16.40 -> 31.14 | Improved |
| Replace supervised learning with unsupervised score-function method on Gaussian noise | Gaussian sigma=25 w/o A · Kodak | PSNR (dB): 32.44 -> 31.92 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Neighbor2Neighbor — Neighbor2Neighbor: Self-Supervised Denoising From Single Noisy Images (CVPR 2021) · better on 8 of 13 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- IDR: Self-Supervised Image Denoising via Iterative Data Refinement (CVPR 2022 · similarity 0.89)
- Unsupervised Image Restoration Using Partially Linear Denoisers (TPAMI 2021 · similarity 0.88)
- Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · similarity 0.86)
- Sampling Binary Data by Denoising through Score Functions (ICML 2025 · similarity 0.86)
- Masked and Shuffled Blind Spot Denoising for Real-World Images (CVPR 2024 · similarity 0.86)
- Zero-Shot Blind-spot Image Denoising via Implicit Neural Sampling (CVPR 2025 · 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.
