VBS24: Winner of the "Textual Known Item Search" Task

At the Video Browser Showdown 2024, held as part of the MMM 2024 conference, we secured the top position in the "Textual Known Item Search" category. This competition challenged participants to retrieve specific scenes from a vast video database. Competing against the world's best video retrieval systems, our team's victory highlights the precision and speed of our search methodology.

Contribution

• We introduced enhancements to the Vibro video retrieval tool, focusing on optimizing its performance for the Textual Known Item Search task.
• Implemented a novel method for evaluating retrieval models using logged query data from previous competitions, ensuring optimal model selection for both image- and text-based retrieval.
• Made significant UI improvements to streamline the search process, particularly for novice users.
• Developed and integrated new retrieval models that excel in both general-purpose and context-specific searches, improving accuracy across diverse datasets.

Please follow the links for more information about the award or our Video Search System Vibro.