Reno Kriz, David Etter, Alexander Martin, Cameron Carpenter, Debashish Chakraborty, Hannah Recknor, Reihaneh Iranmanesh, Matthew Maciejewski, Kenton Murray, Eugene Yang, Benjamin Van Durme, Aaron Steven White, Andrew Yates, William Walden
The paper introduces MultiVENT-Raw, a benchmark for retrieval and reasoning over raw videos. It consists of 120,000 raw videos and supports two tasks: retrieval and generation. This benchmark facilitates progress in video understanding.
The ability to retrieve and reason over raw videos has significant implications for information retrieval, social media monitoring, and content understanding. It can help in identifying relevant videos and summarizing event-related content. This can be useful in various applications such as news aggregation and event reporting.
Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.