# Dataset Card for CAVE: Commonsense Anomalies in Visual Environments
🏠 [Project Page](https://smontariol.github.io/cave-visual-anomalies/)
📄 [Paper (EMNLP 2025)](https://aclanthology.org/2025.emnlp-main.1379/)
## Dataset Details
### Dataset Description
CAVE is the first benchmark of **real-world visual anomalies** for evaluating Vision-Language Models (VLMs). It is curated from images captured in real-life settings (photographs and screenshots taken by individuals), sourced from Reddit.
The benchmark is grounded in cognitive science literature on how humans detect and resolve anomalies. Each image is annotated with rich, multi-task annotations that support three open-ended tasks (anomaly description, explanation, and justification), one visual grounding task (anomaly localization via bounding boxes), and classification along four dimensions (anomaly category, severity, surprisal, and complexity) that characterize the anomaly.
CAVE reveals that state-of-the-art VLMs struggle su…
