Research map / Image generation / Image restoration
Self-supervised image denoising
Denoise images without clean targets using blind-spot and Noise2Noise-style training.
60 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Image restoration, keyword cluster “image denoising · blur · noise”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Leveling off Several papers from 2025–26 still report on 1 tracked benchmark, but the best result has barely moved since 2024.
10 papers in 2023–24 and 13 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic held steady, from 3.2% to 3.0%.
Open questions the data points to
- Papers report BM3D on SIDD (PSNR (dB)) anywhere from 25.65 to 37.23: comparisons against it depend on whose number you take.
What looks settled
- SIDD (PSNR): best up to 2024 40.07, best in 2025–26 38.97 from 2 papers, at most 0.2% better, if at all.
No longer reported by papers here: DND (PSNR, last in 2022); DND (SSIM, last in 2023).
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Deep Gaussian Conditional Random Field Network: A Model-Based Deep Network for Discriminative Denoising (CVPR 2016 · 74 citations)
- 2017 · Non-Local Color Image Denoising With Convolutional Neural Networks (CVPR 2017 · 387 citations)
- 2018 · Burst Denoising With Kernel Prediction Networks (CVPR 2018 · 435 citations)
- 2019 · Noise2Void - Learning Denoising From Single Noisy Images (CVPR 2019 · 1,426 citations)
- 2020 · Self2Self With Dropout: Learning Self-Supervised Denoising From Single Image (CVPR 2020 · 437 citations)
- 2021 · Neighbor2Neighbor: Self-Supervised Denoising From Single Noisy Images (CVPR 2021 · 446 citations)
- 2022 · IDR: Self-Supervised Image Denoising via Iterative Data Refinement (CVPR 2022 · 83 citations)
- 2023 · Zero-Shot Noise2Noise: Efficient Image Denoising Without Any Data (CVPR 2023 · 172 citations)
- 2024 · Robust Image Denoising through Adversarial Frequency Mixup (CVPR 2024 · 18 citations)
- 2025 · Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising (CVPR 2025 · 5 citations)
- 2026 · Trainable Nonexpansive Denoisers for Contractive Image Reconstruction (ICML 2026 · 0 citations)
Earliest papers here
- Deep Gaussian Conditional Random Field Network: A Model-Based Deep Network for Discriminative Denoising (CVPR 2016 · 74 citations)
- Non-Local Color Image Denoising With Convolutional Neural Networks (CVPR 2017 · 387 citations)
- Learning Fully Convolutional Networks for Iterative Non-Blind Deconvolution (CVPR 2017 · 200 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
DND · PSNR (higher is better)
6 papers, 2019–2022 · not reported since 2022
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 38.06 | Toward Convolutional Blind Denoising of Real Photographs (CVPR 2019 · 1,193 citations) |
| 2021 | 40.25 | Learning to Generate Realistic Noisy Images via Pixel-level Noise-aware Adversarial Training (NeurIPS 2021 · 35 citations) |
SIDD Benchmark · PSNR (dB) (higher is better)
6 papers, 2021–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 50.76 | Neighbor2Neighbor: Self-Supervised Denoising From Single Noisy Images (CVPR 2021 · 446 citations) |
SIDD validation · PSNR (dB) (higher is better)
6 papers, 2021–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 51.39 | Neighbor2Neighbor: Self-Supervised Denoising From Single Noisy Images (CVPR 2021 · 446 citations) |
DND · SSIM (higher is better)
5 papers, 2019–2023 · not reported since 2023
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 0.942 | Toward Convolutional Blind Denoising of Real Photographs (CVPR 2019 · 1,193 citations) |
| 2021 | 0.955 | NBNet: Noise Basis Learning for Image Denoising With Subspace Projection (CVPR 2021 · 235 citations) |
| 2022 | 0.959 | Generative Adaptive Convolutions for Real-World Noisy Image Denoising (AAAI 2022 · 34 citations) |
Nam · PSNR (higher is better)
5 papers, 2019–2025 · one report since 2024, not better
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 41.31 | Toward Convolutional Blind Denoising of Real Photographs (CVPR 2019 · 1,193 citations) |
| 2022 | 43.52 | Multi-Scale Adaptive Network for Single Image Denoising (NeurIPS 2022 · 17 citations) |
SIDD · PSNR (higher is better)
5 papers, 2021–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 40.07 | Learning to Generate Realistic Noisy Images via Pixel-level Noise-aware Adversarial Training (NeurIPS 2021 · 35 citations) |
What has been tried
51 papers here report 535 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 8%, hurt 22%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Train DnCNN with random frequency mixup instead of adversarial mask generation | OOD Avg. · PSNR (dB) | 37.81 -> 38.17 | Helped | Robust Image Denoising through Adversarial Frequency Mixup (CVPR 2024 · 18 citations) |
| Replace DCS with Noise2Noise supervision | CC and SIDD · PSNR | 37.92 -> 34.08; 35.24 -> 25.42 | Hurt | Positive2Negative: Breaking the Information-Lossy Barrier in Self-Supervised Single Image Denoising (CVPR 2025 · 5 citations) |
| Replace DQN with RA-DQN optimization | MIDD (l=50) · convergence loss | 2.70 -> 1.78 | Helped | All-Optical Nonlinear Diffractive Deep Network for Ultrafast Image Denoising (CVPR 2025 · 3 citations) |
| Add APR sampling to AT-BSN training | SIDD Validation · PSNR (dB) | 37.80 -> 37.23 | Hurt | APR-RD: Complemental Two Steps for Self-Supervised Real Image Denoising (AAAI 2025 · 2 citations) |
| Replace the convolutional encoder-decoder with an MLP | Set11 AWGN, σ=25 and 50 · PSNR (dB) | 28.93 -> 29.52; 25.58 -> 25.89 | Helped | Zero-shot Denoising via Neural Compression: Theoretical and algorithmic framework (NeurIPS 2025 · 1 citation) |
| use pre-trained weights rather than training from scratch | CSet, Gaussian σ=25 · PSNR | 30.90/0.811 -> 31.61/0.841 | Helped | Masked Pre-training Enables Universal Zero-shot Denoiser (NeurIPS 2024 · 1 citation) |
| Add GRD-Net after MBD-Net | Gaussian σ = 25 · PSNR (dB) | 29.39 -> 29.56 | Helped | Zero-Shot Noise2Mean: Gap Minimization for Efficient Denoising from a Single Noisy Image (AAAI 2025 · 0 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| BM3D | SIDD | PSNR (dB) | 25.65 – 37.23 | 2 papers, e.g. AAAI 2020 and AAAI 2025 |
| BM3D | Kodak24, Gaussian σ = 25 | PSNR (dB) | 29.02 – 30.95 | 2 papers, e.g. CVPR 2023 and AAAI 2025 |
What reviewers will expect
Datasets and baselines named in the 6 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- PolyU · 3 of 6 papers
- SIDD · 3 of 6 papers
- Kodak24 · 2 of 6 papers
Baselines
- DIP · 3 of 6 papers
- Self2Self · 3 of 6 papers
- BM3D · 3 of 6 papers
- ZS-N2N · 3 of 6 papers
- CVF-SID · 2 of 6 papers
- DnCNN · 2 of 6 papers
- MASH · 2 of 6 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 23 papers; from OpenAlex and the papers' affiliations).
- Seoul National University · 2 papers
Use it
Download BibTeX of 36 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Image deblurring (73 papers)
- Raw and low-light noise modeling (53 papers)
- Diffusion priors for denoising and deblurring (41 papers)
Elsewhere in the topic
- Dehazing and deraining (118 papers)
- Efficient restoration architectures (113 papers)
- Inverse imaging with learned priors (109 papers)
- Image harmonization and color transformation (102 papers)
- Reflection removal (95 papers)
- Pansharpening and spectral fusion (84 papers)
- Low-light image enhancement (72 papers)
- Learned image compression (69 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).