The YOLO-RD approach introduces a novel way to incorporate dataset-wide knowledge into YOLO models, overcoming limitations of focusing only on the current image.

• Introduces the Retriever-Dictionary (RD) module that retrieves relevant information from the entire dataset
• Enables YOLO models to make more informed detections by accessing broader contextual information
• Improves accuracy while maintaining YOLO's speed advantages
• Particularly valuable for security applications where object detection reliability is critical for surveillance and threat detection

This research matters for security because it enhances the reliability of automated visual monitoring systems, reducing false positives and improving threat detection in complex environments.

YOLO-RD: Introducing Relevant and Compact Explicit Knowledge to YOLO by Retriever-Dictionary