AI RESEARCH
Follow the Saliency: Supervised Saliency for Retrieval-augmented Dense Video Captioning
arXiv CS.CV
•
ArXi:2603.11460v1 Announce Type: new Existing retrieval-augmented approaches for Dense Video Captioning (DVC) often fail to achieve accurate temporal segmentation aligned with true event boundaries, as they rely on heuristic strategies that overlook ground truth event boundaries. The proposed framework, \textbf{STaRC}, overcomes this limitation by supervising frame-level saliency through a highlight detection module. Note that the highlight detection module is trained on binary labels derived directly from DVC ground truth annotations without the need for additional annotation.