Contamination MapVision foundation models
Which benchmark datasets could each model have seen during training? Every coloured square links to a verbatim quote from the model's paper, model card or repository. Grey means not enough has been disclosed to tell.
Contamination map
- TestTest data seen18
- SeenSeen, split unclear7
- TrainTraining data seen2
- Related data seen0
- NoNot seen33
- ?Unknown236
Rows are models, grouped by family; columns are benchmark datasets. Hover a square for the reason, click it for the evidence. Empty outlines mark benchmarks a model cannot take (e.g. image benchmarks for text-only models).
Benchmarks
-
Industrial anomaly detection
8 datasets
Models
| Model | Organisation | Released | Disclosure | Determined |
|---|---|---|---|---|
| AA-CLIP, trained on MVTec AD (for VisA results) | Zhejiang University et al. | 2025-03-09 | Full list | 12% |
| AA-CLIP, trained on VisA | Zhejiang University et al. | 2025-03-09 | Full list | 12% |
| AdaCLIP, trained on all 14 datasets (demo model) | Huazhong University of Science and Technology et al. | 2024-07-22 | Full list | 50% |
| AdaCLIP, trained on MVTec AD + ColonDB | Huazhong University of Science and Technology et al. | 2024-07-22 | Full list | 12% |
| AdaCLIP, trained on VisA + ClinicDB | Huazhong University of Science and Technology et al. | 2024-07-22 | Full list | 12% |
| AdaptCLIP, trained on MVTec AD | Tencent YouTu Lab et al. | 2025-05-15 | Full list | 12% |
| AdaptCLIP, trained on VisA | Tencent YouTu Lab et al. | 2025-05-15 | Full list | 12% |
| Anomaly-OneVision (Anomaly-OV) | Johns Hopkins University / Honda Research Institute USA | 2025-02-11 | Partial list | 25% |
| AnomalyCLIP, trained on MVTec AD | Zhejiang University / Singapore Management University et al. | 2023-10-29 | Full list | 12% |
| AnomalyCLIP, trained on VisA | Zhejiang University / Singapore Management University et al. | 2023-10-29 | Full list | 12% |
| AnomalyDINO | University of Tuebingen et al. (Damm et al.) | 2024-05-23 | Full list | 0% |
| AnomalyGPT, supervised (demo model) | Institute of Automation, Chinese Academy of Sciences / Objecteye / Wuhan AI Research | 2023-08-29 | Partial list | 37% |
| AnomalyGPT, unsupervised on MVTec AD | Institute of Automation, Chinese Academy of Sciences / Objecteye / Wuhan AI Research | 2023-08-29 | Full list | 12% |
| AnomalyGPT, unsupervised on VisA | Institute of Automation, Chinese Academy of Sciences / Objecteye / Wuhan AI Research | 2023-08-29 | Full list | 12% |
| APRIL-GAN (VAND 2023), trained on MVTec AD | Zhejiang University et al. (Xuhai Chen, Yue Han, Jiangning Zhang) | 2023-05-27 | Full list | 12% |
| APRIL-GAN (VAND 2023), trained on VisA | Zhejiang University et al. (Xuhai Chen, Yue Han, Jiangning Zhang) | 2023-05-27 | Full list | 12% |
| OpenAI CLIP | OpenAI | 2021-02-26 | Partial list | 0% |
| DINOv2 | Meta AI | 2023-04-14 | Partial list | 0% |
| DINOv3 (web / LVD-1689M models) | Meta AI | 2025-08-13 | Partial list | 0% |
| FiLo, trained on MVTec AD | Institute of Automation, Chinese Academy of Sciences et al. | 2024-04-21 | Full list | 12% |
| FiLo, trained on VisA | Institute of Automation, Chinese Academy of Sciences et al. | 2024-04-21 | Full list | 12% |
| InCTRL, trained on MVTec AD | Singapore Management University (Jiawen Zhu, Guansong Pang) | 2024-03-11 | Full list | 62% |
| InCTRL, trained on VisA | Singapore Management University (Jiawen Zhu, Guansong Pang) | 2024-03-11 | Full list | 50% |
| MetaUAS | Tencent YouTu Lab (Bin-Bin Gao) | 2025-05-14 | Full list | 0% |
| MuSc | Huazhong University of Science and Technology et al. | 2024-01-30 | Full list | 0% |
| OpenCLIP (LAION-2B checkpoints) | LAION / OpenCLIP (mlfoundations) | 2022-09-14 | Full list | 50% |
| OpenCLIP (LAION-400M checkpoints) | LAION / OpenCLIP (mlfoundations) | 2022-04-01 | Full list | 50% |
| WideResNet-50-2 (supervised ImageNet-1k, torchvision) | PyTorch / torchvision | — | Full list | 100% |
| Segment Any Anomaly (SAA / SAA+) | Huazhong University of Science and Technology et al. | 2023-05-18 | Full list | 0% |
| SAM 2 / SAM 2.1 | Meta AI | 2024-07-29 | Partial list | 0% |
| Segment Anything Model (SAM) | Meta AI | 2023-04-05 | Partial list | 25% |
| SigLIP | Google DeepMind | 2023-03-27 | Partial list | 25% |
| SigLIP 2 | Google DeepMind | 2025-02-20 | Partial list | 25% |
| UniVAD | Institute of Automation, Chinese Academy of Sciences et al. | 2024-12-04 | Full list | 0% |
| VCP-CLIP, trained on MVTec-AD | Institute of Automation, Chinese Academy of Sciences et al. | 2024-07-17 | Full list | 12% |
| VCP-CLIP, trained on VisA | Institute of Automation, Chinese Academy of Sciences et al. | 2024-07-17 | Full list | 12% |
| WinCLIP / WinCLIP+ | Amazon AWS AI Labs / University of Wisconsin-Madison | 2023-03-26 | Full list | 50% |