AIGeeksGroup/MoBench

# MoBen Dataset The benchmark dataset covering two tasks: **Music-to-Dance (M2D)** and **Text-to-Motion (T2M)**. ## Structure ``` MoBen/ ├── M2D/ # 1,500 music-dance pairs (self-contained) │ ├── aistpp/ # 500 samples from AIST++ │ ├── finedance/ # 500 samples from FineDance │ └── popdanceset/ # 500 samples from PopDanceSet │ ├── T2M/ # Pose-level text annotations (body + hand) │ ├── body_texts/ # 14 sub-sources, body/torso/leg descriptions │ └── hand_texts/ # 14 sub-sources, finger-level descriptions │ └── T2M_pathref.txt # 10,000 selected motion paths (reference only) ``` ## Music-to-Dance (M2D) Each sub-dataset contains `motions/*.pkl` and `wavs/*.wav`. Motion and audio files are matched by filename. | Source | Motions | Audio | |---|---|---| | AIST++ | 500 | 56 (shared across clips) | | FineDance | 500 | 500 | | PopDanceSet | 500 | 500 |