| # 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 | | |