Research map / Image generation / Fast diffusion sampling
Diffusion models as perception tools
Use diffusion models for segmentation, adaptation, and data generation in discriminative tasks.
198 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Fast diffusion sampling, keyword cluster “score · distribution · denoising diffusion”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
80 papers in 2023–24 and 109 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 17.4% to 8.2%.
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
- 2018 · COSINE: Community-Preserving Social Network Embedding From Information Diffusion Cascades (AAAI 2018 · 67 citations)
- 2019 · Diffusion Scattering Transforms on Graphs (ICLR 2019 · 30 citations)
- 2020 · Pattern-Structure Diffusion for Multi-Task Learning (CVPR 2020 · 93 citations)
- 2022 · MS-HGAT: Memory-Enhanced Sequential Hypergraph Attention Network for Information Diffusion Prediction (AAAI 2022 · 80 citations)
- 2023 · Ambiguous Medical Image Segmentation Using Diffusion Models (CVPR 2023 · 210 citations)
- 2024 · MedSegDiff-V2: Diffusion-Based Medical Image Segmentation with Transformer (AAAI 2024 · 263 citations)
- 2025 · Boosting Consistency in Story Visualization with Rich-Contextual Conditional Diffusion Models (AAAI 2025 · 34 citations)
- 2026 · Directing Uncertainty-Aware Information Flow for Robust Diffusion Prediction (AAAI 2026 · 2 citations)
Earliest papers here
- COSINE: Community-Preserving Social Network Embedding From Information Diffusion Cascades (AAAI 2018 · 67 citations)
- Few-shot Autoregressive Density Estimation: Towards Learning to Learn Distributions (ICLR 2018 · 39 citations)
- Diffusion Scattering Transforms on Graphs (ICLR 2019 · 30 citations)
What has been tried
151 papers here report 1,623 measured changes (a component added, removed or swapped, compared with a baseline): helped 74%, no clear effect 5%, hurt 17%, 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 |
|---|---|---|---|---|
| replace spatial attention with U-SA in Anchor Condition | AMOS · average Dice | 83.5 -> 86.7 | Helped | MedSegDiff-V2: Diffusion-Based Medical Image Segmentation with Transformer (AAAI 2024 · 263 citations) |
| Add backward universal guidance to forward guidance | Object-location-guided image generation · object category and location match | Helped | Universal Guidance for Diffusion Models (ICLR 2024 · 148 citations) | |
| Add Spatial Anomaly Embedding, masked diffusion loss, and adaptive attention re-weighting | MVTec pixel-level localization · AUROC | 81.3 -> 99.1 | Helped | AnomalyDiffusion: Few-Shot Anomaly Image Generation with Diffusion Model (AAAI 2024 · 140 citations) |
| Randomize insertion timestep t0 rather than fixing it at 0.5 | Few-shot classification, all domains tested in the intensity ablation · average classification accuracy | 51% | Helped | Effective Data Augmentation With Diffusion Models (ICLR 2024 · 78 citations) |
| Add the discriminator to adaptive resampling | PSV hard mask · PSNR | 25.47 -> 25.94 | Helped | Structure Matters: Tackling the Semantic Discrepancy in Diffusion Models for Image Inpainting (CVPR 2024 · 76 citations) |
| add fractal blending to original images | Stanford Cars · Top-1 accuracy (%) | 85.52 -> 86.73 | Helped | DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models (CVPR 2024 · 68 citations) |
| Replace ViT scene embeddings with one-hot scene-label embeddings | NYU Depth v2 · RMSE | 0.218 -> 0.244 | Hurt | ECoDepth: Effective Conditioning of Diffusion Models for Monocular Depth Estimation (CVPR 2024 · 52 citations) |
| Use binary label encoding instead of one-hot encoding | iSAID, R = 1 · IoU | 47.98 -> 51.00 | Helped | SatSynth: Augmenting Image-Mask Pairs through Diffusion Models for Aerial Semantic Segmentation (CVPR 2024 · 51 citations) |
| Add channel-wise attention to MLP + TSConv | THINGS-EEG retrieval · Top-1 | 21.65± 6.22 -> 28.64±6.39 | Helped | Visual Decoding and Reconstruction via EEG Embeddings with Guided Diffusion (NeurIPS 2024 · 38 citations) |
| add image-level condition | FlintstonesSV · Char-Acc | 56.44 -> 61.88 | Helped | Boosting Consistency in Story Visualization with Rich-Contextual Conditional Diffusion Models (AAAI 2025 · 34 citations) |
| Use a 3D ControlNet instead of stacking images in channels with a 2D ControlNet | Texas Housing, t′ > t · LPIPS | 0.3874 -> 0.3247 | Helped | DiffusionSat: A Generative Foundation Model for Satellite Imagery (ICLR 2024 · 31 citations) |
| Add the dual-branch Task-Plugin | Desnowing · FID | 34.50 -> 34.30 | Helped | Diff-Plugin: Revitalizing Details for Diffusion-based Low-level Tasks (CVPR 2024 · 29 citations) |
What reviewers will expect
Datasets and baselines named in the 73 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 6 of 73 papers
- ImageNet-1K · 3 of 73 papers
- MNIST · 3 of 73 papers
- CIFAR-100 · 2 of 73 papers
- Twitter · 2 of 73 papers
- PANDA · 2 of 73 papers
- KuaiRec · 2 of 73 papers
- PanNuke · 2 of 73 papers
Baselines
- UNet++ · 3 of 73 papers
- TabDDPM · 2 of 73 papers
- SEISMIC · 2 of 73 papers
- DeepHawkes · 2 of 73 papers
- CasCN · 2 of 73 papers
- CTCP · 2 of 73 papers
- NGCF · 2 of 73 papers
- LightGCN · 2 of 73 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 189 papers; from OpenAlex and the papers' affiliations).
- Tsinghua University · 8 papers
- Zhejiang University · 8 papers
- Carnegie Mellon University · 7 papers
- Shanghai Jiao Tong University · 6 papers
- Stanford University · 6 papers
- University of Chinese Academy of Sciences · 6 papers
- Chinese Academy of Sciences · 6 papers
- Tencent · 5 papers
- University of Science and Technology of China · 5 papers
- The Chinese University of Hong Kong · 5 papers
Use it
Download BibTeX of 29 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
- Diffusion theory and design space (252 papers)
- Distributional control of diffusion (109 papers)
- Denoising diffusion probabilistic models (97 papers)
- Guidance for conditional diffusion (123 papers)
- Diffusion posterior sampling (72 papers)
- Diffusion theory and learned samplers (121 papers)
Elsewhere in the topic
- Training-free diffusion acceleration (257 papers)
- Masked diffusion language model decoding (147 papers)
- Fast diffusion ODE solvers (120 papers)
- Flow matching distillation (95 papers)
- Distillation for few-step generation (84 papers)
- Rectified flow straightening (70 papers)
- Discrete diffusion solvers (64 papers)
- Guidance and alignment for flow matching (34 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).