Query-Focused Video Summarization Dataset

本数据集专为查询焦点视频摘要（Query-Focused Video Summarization）设计，旨在解决用户对视频摘要的主观偏好问题。数据集基于UT Egocentric (UTE)数据集构建，包含长达3-5小时的日常场景视频，覆盖多种事件。数据集通过密集的每视频镜头概念注释，使用二进制语义向量表示视频镜头中的语义信息。创建过程中，利用Amazon Mechanical Turk进行概念注释，同时招募学生志愿者进行视频摘要的标注。该数据集特别适用于开发和评估能够根据用户查询生成个性化视频摘要的算法，旨在通过保留视频的语义信息，提供更符合用户需求的视频浏览体验。This dataset is specifically designed for Query-Focused Video Summarization, aiming to address the subjective preference issues of users regarding video summarization. Built upon the UT Egocentric (UTE) dataset, it contains daily-scene videos covering a total duration of 3 to 5 hours, spanning various types of events. The dataset features dense per-video shot concept annotations, where binary semantic vectors are employed to represent the semantic information within each video shot. During its construction, Amazon Mechanical Turk was utilized for concept annotation, while student volunteers were recruited to label video summaries. This dataset is particularly well-suited for developing and evaluating algorithms that can generate personalized video summaries based on user queries, with the goal of providing a video browsing experience that better aligns with user needs by preserving the semantic information of the videos.