QVHighlights

基于查询的视频亮点 (QVHighlights) 数据集是一个数据集，用于从给定自然语言 (NL) 的视频中检测自定义时刻和亮点。它包含超过 10,000 个 YouTube 视频，涵盖了广泛的主题，从生活方式 vlog 视频中的日常活动和旅行到新闻视频中的社会和政治活动。数据集中的每个视频都标注有：（1）人工编写的自由形式的 NL 查询，（2）视频中的相关时刻 w.r.t。查询，以及 (3) 所有查询相关剪辑的五点量表显着性分数。

The Query-based Video Highlights (QVHighlights) dataset is a benchmark dataset designed for detecting customized moments and highlights from videos given natural language (NL) queries. It includes over 10,000 YouTube videos covering a wide range of topics, spanning from daily activities and travel in lifestyle vlog videos to social and political events in news videos. Each video in the dataset is annotated with three types of content: (1) manually written free-form natural language queries, (2) relevant moments within the video with respect to the corresponding query, and (3) five-point scale significance scores for all query-related clips.

搜集汇总

背景与挑战

背景概述

QVHighlights是一个用于基于自然语言查询检测视频中自定义时刻和亮点的数据集，包含超过10,000个YouTube视频，覆盖从日常生活到新闻的多样主题。每个视频都标注了人工编写的查询、相关时刻和显着性分数，支持视频理解任务。