Fudan-FUXI/LiFT-HRA-20K
收藏官方服务：
资源简介：
LiFT-HRA是一个高质量的人类偏好标注数据集，用于训练视频文本到文本的奖励模型。该数据集中的所有视频分辨率至少为512×512。
LiFT-HRA is a high-quality Human Preference Annotation dataset that can be used to train video-text-to-text reward models. All videos in the LiFT-HRA dataset have resolutions of at least 512×512. This dataset is designed to help align text-to-video models.
提供机构：
Fudan-FUXI
