Existing weakly supervised video anomaly detection methods mainly employ Multiple Instance Learning (MIL) to identify abnormal snippets in untrimmed videos. However, the semantics and presentations of ...