RoMo-HML-263 is the RoMo body corpus packed in the 263-dimensional HumanML3D motion-feature representation, paired with rich multi-level text descriptions. It is the drop-in companion for training and evaluating models built around the HumanML3D feature set, sized at the RoMo scale (~815K clips).
⚠️ Access: This dataset is currently private / internal. It will be released publicly in conjunction with the RoMo paper.… See the full description on the dataset page:
https://huggingface.co/datasets/RoMoDataset/RoMo-HML-263.