RAPID introduces a retrieval-augmented approach that dramatically improves text-based video event search by addressing the critical context gap in existing systems.

• Leverages parallel inference drafting to enrich queries with missing contextual details
• Significantly improves retrieval accuracy even with ambiguous or incomplete search queries
• Enhances processing efficiency while maintaining high-quality results
• Creates a more robust foundation for security monitoring systems

This research is vital for security applications where identifying specific events in vast video archives quickly and accurately can be mission-critical for threat detection and incident response.

RAPID: Retrieval-Augmented Parallel Inference Drafting for Text-Based Video Event Retrieval