HiREST

该数据集名为HiREST，包含总共3,400组文本视频对，平均视频时长为287秒，总时长达到270小时。在这3,400个视频中，有1,800个视频可以被剪辑成一个时刻。平均时刻长度为148秒，占原始视频长度的55%。此外，该数据集是按照查询而非视频进行划分的，因此训练/验证/测试集的查询数量分别为546、292和546。该数据集的规模为3,400组文本视频对，总时长为270小时，其任务是时刻检索。

This dataset is named HiREST, which contains a total of 3,400 text-video pairs. The average duration of individual videos is 287 seconds, with an overall total duration of 270 hours. Out of the 3,400 videos, 1,800 can be clipped into moments, with an average moment length of 148 seconds, accounting for 55% of the original video's duration. Furthermore, this dataset is split by query rather than by video. The numbers of queries in the training, validation, and test splits are 546, 292, and 546 respectively. With 3,400 text-video pairs and a total duration of 270 hours, this dataset is designed for the moment retrieval task.