fudan-generative-ai/hallo3_training_data

该数据集是用于训练Hallo3模型的训练数据，包含超过70小时专注于说话人头部的纯净视频和超过50个涵盖各种现实世界场景的野生场景视频片段。数据集适用于学术研究和非商业模型训练。

This dataset is the training data for the Hallo3 model, including over 70 hours of pure talking-head videos focusing on the speakers face and speech, and more than 50 wild-scene video clips covering a variety of real-world scenarios. It is suitable for academic research and non-commercial model training.