Takeaways from the SCALE 2024 Workshop on Video-based Event Retrieval

Explore key insights from the SCALE 2024 Workshop on Video-based Event Retrieval in this comprehensive talk presented by Reno Kriz and Kate Sanders from the Center for Language & Speech Processing at Johns Hopkins University. Delve into the challenges of multilingual event-centric video retrieval, focusing on user-generated content from non-professional sources. Learn about the development of MultiVENT 2.0, a large-scale video retrieval dataset, and the efforts of five sub-teams working on improving models for various modalities including Vision, Optical Character Recognition (OCR), Audio, and Text. Discover the three primary findings: the importance of extracting specific text from videos, the benefits of LLM summarization for noisy text outputs, and the necessity of fusing multiple modalities for optimal performance. Gain valuable insights into the evolving landscape of event analysis and retrieval in the age of user-generated content.