QVHighlights

该数据集名为QVHighlights，是最近公布的集时刻检索和高亮检测于一体的数据集，它为这两项任务都提供了标注信息。对于测试集的评估，只能通过将预测结果提交至QVHighlights服务器来进行。该数据集规模宏大，包含超过10,000个视频，每个视频都配有由人类编写的文本查询。任务的范畴涉及时刻检索和高亮检测。

The dataset named QVHighlights, a recently released resource, integrates both moment retrieval and highlight detection tasks, and provides annotated data for both tasks. For test set evaluation, prediction results can only be submitted to the QVHighlights server for official assessment. This is a large-scale dataset containing over 10,000 videos, each paired with human-written text queries. The task scope covers moment retrieval and highlight detection.

搜集汇总

背景与挑战

背景概述

QVHighlights数据集专注于通过自然语言查询检测视频中的关键时刻和高光片段，由UNC Chapel Hill的研究团队组织，并提供在线竞赛平台供参与者提交模型结果。