Picaa-AI/Lulu

生成的IP视频数据集，包含生成视频和参考视频两种类型。通过视频的metadata信息区分类型，其中generated为生成视频，reference为参考视频。数据集中每个视频都有对应的IP形象名称（ip_name），只有名称相同的视频可以进行相互比较。该数据集旨在通过无参考或有参考的方法，区分生成视频中质量不佳的样本，这些样本可能存在执行指令错误、形象特征不正确或画面不符合物理规律等问题。区分方法可以是排序或分类等。

Generated IP Videos dataset, which includes two types of videos: generated and reference. The type of each video is distinguished by its metadata, with generated being the generated videos and reference being the reference videos. Each video in the dataset has an associated IP character name (ip_name), and only videos with the same name can be compared. The purpose of this dataset is to distinguish poor-quality samples in the generated videos using reference-free or reference-based methods, which may have issues such as incorrect instruction following, incorrect character features, or visuals that do not conform to physical laws. The distinction can be in the form of sorting or classification, etc.