MovieNet

该数据集名为MovieNet，包含了1100部电影，拥有大量的多模态数据和注释，总时长约3000小时，是迄今为止用于电影理解分析的最大数据集。该数据集被划分为训练集、验证集和测试集，分别包含660部电影、220部电影和220部电影。此外，该数据集为190部训练视频、64部验证视频和64部测试视频提供了场景边界信息。其规模之大，任务之重点是视频场景分割。

This dataset, named MovieNet, consists of 1100 films, featuring abundant multimodal data and annotations, with a total duration of approximately 3000 hours, making it the largest dataset for film understanding and analysis to date. It is divided into training, validation and test sets, which contain 660, 220 and 220 films respectively. Additionally, this dataset provides scene boundary information for 190 training videos, 64 validation videos and 64 test videos. Given its large scale, the core task of this dataset focuses on video scene segmentation.