zachz/VideoCon-PC

VideoCon-PC是一个视频-语言对分类数据集，来源于VideoCon（Bansal et al., 2023）。每个样本包含一个视频和一个字幕。标签为1表示字幕正确描述了视频，标签为0表示字幕是语义上合理的对比字幕（实体/动作/属性交换，事件顺序翻转）。数据集包含1138个样本，来自ActivityNet的569个视频（移除了2个损坏的视频）。数据集的特征包括视频、字幕和标签，任务类别为视频分类，语言为英语。VideoCon-PC is a video-language pair classification dataset from VideoCon (Bansal et al., 2023). Each row contains a video and a caption. Label=1 means the caption correctly describes the video; label=0 means it is a semantically-plausible contrast caption (entity/action/attribute swaps, event order flips). Source: videocon/videocon (videocon_human.csv) — 569 videos from ActivityNet, 1138 pairs total (2 corrupt videos removed).