tomg-group-umd/argus

ARGUS数据集包含视频片段名称和对应的人类编写的字幕。这个数据集被用来评估视频字幕中的虚构内容（ArgusCost-H）和遗漏内容（ArgusCost-O）的程度。数据集由训练集组成，视频片段来自于现有的视频理解数据集、文本到视频生成数据集以及遵循创意共享许可证的公开可用视频源。

The ARGUS dataset consists of video clip names and corresponding human-written captions. This dataset is used to evaluate the degree of hallucinated content (ArgusCost-H) and omitted content (ArgusCost-O) in video captions. The dataset is composed of a training set, with video clips sourced from existing video understanding datasets, text-to-video generation datasets, and publicly available videos under Creative Commons licenses.