Video Summarization Leveraging Multimodal Information for Presentations

This demonstration introduces a video summarization system, leveraging multimodal information to efficiently extract essential contents from presentations. In contrast to existing methods focusing primarily on daily life videos and solely utilizing visual information, our system extracts multimodal information, including speech, text, and visual information from videos of presentations. Specifically, the proposed system extracts crucial slide texts from key-frames as queries to filter speech transcripts. By piecing together the video clips corresponding to the filtered speech transcripts, our system outputs the final video summarizations. The evaluation on ICCV 2017 videos demonstrates the effectiveness of the proposed system compared with the lead-3 baseline.