BrokenVideos

该数据集是一个包含3,254个人工智能生成的视频的基准数据集，每个视频都配有精心标注的像素级遮罩，这些遮罩突出了视觉损坏的区域。该数据集涵盖了多样的场景类别和生成模型来源，并采用了一套全面的标注协议，以便对细微的瑕疵进行精确识别。规模上，该数据集包含3,254个高分辨率的人工智能生成视频，约33.6万个高质量的手动瑕疵区域标注。其任务是对人工智能生成视频中的瑕疵进行定位和检测。

This dataset is a benchmark dataset encompassing 3,254 artificially generated videos, each paired with meticulously annotated pixel-level masks that highlight regions of visual artifacts. This dataset covers diverse scene categories and generative model sources, and employs a comprehensive annotation protocol to enable precise identification of subtle artifacts. In terms of scale, this dataset includes 3,254 high-resolution artificially generated videos, with approximately 336,000 high-quality manual annotations for artifact regions. The task of this dataset is to localize and detect artifacts in artificially generated videos.