Weakly Supervised Video Anomaly Detection

Weakly Supervised Video Anomaly Detection (WS-VAD) refers to the use of models trained primarily on video-level labels to identify abnormal behaviors in video sequences without the need for explicit frame-level annotations. This approach significantly reduces annotation costs by utilizing coarse labels, with its core challenge lying in accurately localizing anomalies temporally and effectively distinguishing subtle abnormal activities from normal background events using limited supervisory signals. WS-VAD holds significant application value in areas such as surveillance, security, and healthcare.