本数据集用于 AI 生成视频的破损/异常区域检测与分割（broken video artifact detection & segmentation）。
数据来源于多种主流视频生成模型，包含原始视频、逐帧图像，以及逐像素的破损实例标签。
The dataset is for detecting and segmenting broken / artifact regions in AI-generated videos.
It covers several mainstream video-generation models and provides source videos, per-frame images, and per-pixel broken-instance labels.
Easyanimatev4, OpenSora1.2, kling1.5, luma1.6, minimax, tongyi
（部分样本 ID… See the full description on the dataset page:
https://huggingface.co/datasets/Luuu213u/broken_video_dataset.
