Researchers have introduced USV, a new dataset comprising approximately 224,000 user-generated short-form videos. This dataset is designed to advance the understanding of high-level semantic information in videos, moving beyond instance-level recognition. To facilitate research, the paper also establishes topic recognition and video-text retrieval tasks on USV, proposing baseline methods like MMF-Net and VTCL. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Introduces a new dataset and baseline methods to advance research in understanding user-generated short-form videos.
RANK_REASON The cluster contains an academic paper introducing a new dataset and methods for video understanding. [lever_c_demoted from research: ic=1 ai=1.0]
