Anomaly detection attempts to identify “unusual” instances, that is, instances that deviate considerably from what is expected. Machine learning has become pretty good at detecting such anomalies—as long as abundant data are available. If data are scarce, however, the problem remains challenging. State-of-the-art approaches are multi-modal: they use several types of data simultaneously, usually images and text. But humans can often do well at anomaly detection using images alone. Our recent paper “AnomalyDINO: Boosting Patch-based Few-shot Anomaly Detection with DINOv2” has shown that the same is true also for machine learning. The reviewers agreed, and so our paper has been accepted for WACV 2025. 😁 Congrats to Simon and Mike, and thanks to our collaborator Asja! 🍷🍷🍷🍷