RoMoDataset/RoMo-SMPLX

RoMo-SMPLX 是一个大规模数据集，包含约 787,000 个单人身体运动序列，以 SMPL-X 参数空间表示，并配有多级文本描述（从简短标签到详细段落）。该数据集基于野外互联网视频，通过 GVHMR 进行单目 3D 姿势估计恢复，帧率为 30 fps，使用 SMPL-X 中性身体模型。每个序列包括身体姿势、形状参数、根平移和根方向等原始参数，以及 5 个级别的文本描述。数据集以 Parquet 格式（zstd 压缩分片）存储，分为训练、验证和测试集，适用于文本到 3D 运动生成、身体运动分析等任务。许可证为 CC BY-NC 4.0，仅限非商业使用。RoMo-SMPLX is a large-scale dataset of single-person body motion sequences in the SMPL-X parameter space, paired with multi-level text descriptions (from tags to paragraphs). It contains approximately 787,000 sequences derived from in-the-wild internet videos, with 3D pose estimation via GVHMR for monocular recovery, a frame rate of 30 fps, and uses the SMPL-X neutral body model. Each sequence includes raw parameters such as body pose, shape parameters, root translation, and root orientation, along with five levels of text captions. The dataset is stored in Parquet format (zstd-compressed shards) and split into train, validation, and test sets, suitable for tasks like text-to-3D motion generation and body motion analysis. It is licensed under CC BY-NC 4.0 for non-commercial use only.