Find And Focus: Retrieve And Localize Video Events With Natural Language Queries

The thriving of video sharing services brings new challenges to video retrieval, e.g. the rapid growth in video duration and content diversity. Meeting such challenges calls for new techniques that can effectively retrieve videos with natural language queries. Existing methods along this line, which mostly rely on embedding videos as a whole, remain far from satisfactory for real-world applications due to the limited expressive power. In this work, we aim to move beyond this limitation by delving into the internal structures of both sides, the queries and the videos. Specifically, we propose a new framework called Find and Focus (FIFO), which not only performs top-level matching (paragraph vs. video), but also makes part-level associations, localizing a video clip for each sentence in the query with the help of a focusing guide. These levels are complementary - the top-level matching narrows the search while the part-level localization refines the results. On both ActivityNet Captions and modified LSMDC datasets, the proposed framework achieves remarkable performance gains (Project Page: https://ycxioooong.github.io/projects/fifo).