Research map / Image generation / Autoregressive and latent generation
Reconstruction-based diffusion image detection
Detect diffusion-generated images using reconstruction error and diffusion-based anomaly modeling.
45 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2021–2026) · an approach in Autoregressive and latent generation, keyword cluster “ai · watermarking · generated”
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.
17 papers in 2023–24 and 27 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 3.6% to 3.0%.
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
- 2021 · Improved Contrastive Divergence Training of Energy-Based Models (ICML 2021 · 26 citations)
- 2023 · Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Models (CVPR 2023 · 169 citations)
- 2024 · DiffusionEdge: Diffusion Probabilistic Model for Crisp Edge Detection (AAAI 2024 · 70 citations)
- 2025 · Dual-Interrelated Diffusion Model for Few-Shot Anomaly Image Generation (CVPR 2025 · 26 citations)
- 2026 · Explainable Synthetic Image Detection Through Diffusion Timestep Ensembling (AAAI 2026 · 2 citations)
Earliest papers here
- Improved Contrastive Divergence Training of Energy-Based Models (ICML 2021 · 26 citations)
- Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Models (CVPR 2023 · 169 citations)
- Dream the Impossible: Outlier Imagination with Diffusion Models (NeurIPS 2023 · 14 citations)
What has been tried
34 papers here report 369 measured changes (a component added, removed or swapped, compared with a baseline): helped 67%, no clear effect 8%, hurt 23%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add Uncertainty Distillation without Adaptive FFT-filter | BSDS · ODS | 0.816 -> 0.831 | Helped | DiffusionEdge: Diffusion Probabilistic Model for Crisp Edge Detection (AAAI 2024 · 70 citations) |
| take the minimum reconstruction error across multiple AEs instead of using a single AE | SD1.1, SD1.5, SD2.1, KD2.1, MJ4, MJ5, and MJ5.1 · AP | 0.959; 0.957; 0.991; 0.996; 0.998; 0.993; 0.994 | Helped | AEROBLADE: Training-Free Detection of Latent Diffusion Images Using Autoencoder Reconstruction Error (CVPR 2024 · 53 citations) |
| add ESR using LaRE instead of the image-only baseline | GenImage, average across 8 generators · accuracy | 73.1 -> 81.7 | Helped | LaRE^2: Latent Reconstruction Error Based Method for Diffusion-Generated Image Detection (CVPR 2024 · 48 citations) |
| use Simplex instead of Gaussian noise for DDPM pretraining with the ADE20K encoder | IMD2020 and AutoSplicing · pixel-level F1 | .470 -> .511; .346 -> .507 | Helped | DiffForensics: Leveraging Diffusion Prior to Image Forgery Detection and Localization (CVPR 2024 · 39 citations) |
| add the Background Compensation Module | MVTec bottle, pill, toothbrush · pixel-level AUROC | 98.8 -> 99.1 | Helped | Dual-Interrelated Diffusion Model for Few-Shot Anomaly Image Generation (CVPR 2025 · 26 citations) |
| replace standard diffusion with Discrete Action-Detection Diffusion | THUMOS14 · Avg mAP | 66.0 -> 69.0 | Helped | Action Detection via an Image Diffusion Process (CVPR 2024 · 16 citations) |
| Use direct ˜z0 correction instead of ˜zt−1 correction | MVTec-AD · image-level AUROC | 99.3 -> 99.2 | Hurt | Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · 13 citations) |
| Use preset frequency mask #5 instead of mask #1 | DALL·E 3 · AUC | 58.5 -> 78.3 | Helped | FIRE: Robust Detection of Diffusion-Generated Images via Frequency-Guided Reconstruction Error (CVPR 2025 · 13 citations) |
| Add prompt tuning to the baseline | OpenSDID SD1.5 · Image F1 | 0.7613 -> 0.8576 | Helped | OpenSDI: Spotting Diffusion-Generated Images in the Open World (CVPR 2025 · 11 citations) |
| Replace the 1D statistic with the 6D statistic | 12 pairwise OOD tasks average · AUROC | 0.806 -> 0.850 | Helped | Out-of-Distribution Detection with a Single Unconditional Diffusion Model (NeurIPS 2024 · 10 citations) |
| Increase the number of function evaluations from NFE-1 to NFE-10 | MVTec-AD · AUROCcls | 97.42 -> 99.14 | Helped | A Unified Latent Schrodinger Bridge Diffusion Model for Unsupervised Anomaly Detection and Localization (CVPR 2025 · 9 citations) |
| Add the SG network intermediate block to the diffusion model | MVTec AD · Image AUC (cls) | 0.812 -> 0.834 | Helped | DZAD: Diffusion-based Zero-shot Anomaly Detection (AAAI 2025 · 6 citations) |
What reviewers will expect
Datasets and baselines named in the 16 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- MVTec AD · 4 of 16 papers
- VisA · 3 of 16 papers
- ImageNet · 2 of 16 papers
- CIFAR-100 · 2 of 16 papers
- ImageNet-1k · 2 of 16 papers
- CIFAR-10 · 2 of 16 papers
Baselines
- SimpleNet · 2 of 16 papers
- DRAEM · 2 of 16 papers
- RD4AD · 2 of 16 papers
- SecMI · 2 of 16 papers
- PIA · 2 of 16 papers
- PIAN · 2 of 16 papers
- CNNDet · 2 of 16 papers
- AEROBLADE · 2 of 16 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 44 papers; from OpenAlex and the papers' affiliations).
- Tencent · 3 papers
- Zhejiang University · 3 papers
- Shanghai Jiao Tong University · 3 papers
- National University of Singapore · 2 papers
- Sungkyunkwan University · 2 papers
- LG AI Research · 2 papers
- Sun Yat-sen University · 2 papers
- Georgia Institute of Technology · 2 papers
- Cornell University · 2 papers
- University of Michigan · 2 papers
Use it
Download BibTeX of 25 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
- Generalizable fake image detection (119 papers)
- Robust diffusion watermarking (79 papers)
Elsewhere in the topic
- Classical generative image synthesis (147 papers)
- High-resolution diffusion synthesis (140 papers)
- Retrieval and matching with generative features (140 papers)
- Generative latent space tooling (129 papers)
- Latent diffusion with learned tokenizers (114 papers)
- Synthetic training data from generators (111 papers)
- Compositional scene generation (105 papers)
- Next-scale and token autoregressive images (90 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).