wruisi/ViMoGen-228K

ViMoGen-228K是一个大规模的人类运动生成数据集，包含228,000个高质量的运动样本。数据集整合了高保真的光学运动捕捉数据、网络视频中的语义标注动作以及由最先进的视频生成模型生成的合成样本。数据集提供了文本-动作对和文本-视频-动作三元组，极大地扩展了语义多样性。数据集以JSON格式组织，包含一个统一的注释文件和三个按数据类型分割的注释文件。每个数据样本包含唯一标识符、数据子集、分割信息、运动文本注释、视频文本注释、运动数据文件路径和可选的视频文件路径。运动数据文件是PyTorch序列化的字典，包含运动特征、相机外参和内参矩阵。

ViMoGen-228K is a large-scale dataset for human motion generation, comprising 228,000 high-quality motion samples. The dataset integrates high-fidelity optical MoCap data with semantically annotated motions from web videos and synthesized samples generated by state-of-the-art ViGen models. It includes both text-motion pairs and text-video-motion triplets, substantially expanding semantic diversity. The dataset is organized in JSON format, including one unified annotation file and three additional annotation files split by data type. Each entry contains a unique identifier, data subset, split information, motion text annotation, video text annotation, motion data file path, and optional video file path. The motion data files are PyTorch serialized dictionaries containing motion features, camera extrinsic parameters, and intrinsic matrix.