Sherlock introduces a groundbreaking approach to video surveillance by extracting structured information about abnormal events - identifying what happened, when it occurred, and where it took place.

• Proposes a novel Multi-scene Video Abnormal Event Extraction and Localization (M-VAE) task
• Leverages a spatial-sensitive large language model with global-local processing capability
• Extracts complete event quadruples (subject, event type, object, scene)
• Provides precise localization of security incidents across diverse environments

This research significantly enhances security monitoring systems by moving beyond simple anomaly flagging to comprehensive event understanding, enabling more intelligent and actionable security responses in surveillance applications.

Original Paper: Sherlock: Towards Multi-scene Video Abnormal Event Extraction and Localization via a Global-local Spatial-sensitive LLM