Lan Wang

QSVideo: Query-Conditioned Semantic Temporal Retrieval for Video Understanding

Wei Ao, Lan Wang, Vishnu Naresh Boddeti

The performance of vision-language models (VLMs) in video understanding declines with increasing video duration, as video moments unrelated to the query confuse their language components. Multimodal retrieval has emerged as a critical component of video understanding, addressing…Read more