| --- |
| language: |
| - en |
| license: cc-by-nc-4.0 |
| size_categories: |
| - 10K<n<100K |
| task_categories: |
| - image-to-video |
| - video-generation |
| - computer-vision |
| pretty_name: MACE-Dance |
| --- |
| |
| # 🎵 MACE-Dance Dataset |
|
|
| [**Project Page**](https://macedance.github.io/) | [**Paper**](https://huggingface.co/papers/2512.18181) | [**Code**](https://github.com/AMAP-ML/MACE-Dance) |
|
|
| **MACE-Dance** is a large-scale dataset for **music-driven dance video generation**, released with the paper: |
|
|
| > **MACE-Dance: Motion-Appearance Cascaded Experts for Music-Driven Dance Video Generation** |
|
|
| It is designed to support research on generating dance videos that are both: |
|
|
| - 🕺 **kinematically plausible** |
| - 🎨 **visually coherent** |
| - 🎼 **well aligned with music** |
|
|
| --- |
|
|
| ## ✨ Overview |
|
|
| The dataset (referred to as **MA-Data** in the paper) contains approximately: |
|
|
| - **70K** dance video clips |
| - **5–10 seconds** per clip |
| - **116 hours** in total |
| - **20+ dance genres** |
|
|
| The data is curated from two complementary sources: |
|
|
| ### 1. Motion-centric subset |
| - Derived from **FineDance** |
| - Front-view rendered dance videos from 3D motion sequences |
| - Focused on **professional dance motion quality** |
|
|
| ### 2. Appearance-centric subset |
| - Collected from high-engagement internet dance videos |
| - Focused on **visual appearance diversity and realism** |
|
|
| This design helps benchmark both **motion quality** and **appearance quality** in music-driven dance video generation. |
|
|
| --- |
|
|
| ## 📂 Folder Structure |
|
|
| ```bash |
| MACE-Dance/ |
| ├── Appearance/ |
| └── Kinematic/ |
| ``` |
|
|
| > The exact file organization may vary depending on the released version. |
|
|
| --- |
|
|
|
|
| ## 🧹 Data Curation |
|
|
| For the in-the-wild subset, we apply a multi-stage cleaning pipeline: |
|
|
| - ✂️ shot boundary detection |
| - 🚶 motion filtering |
| - 🧍 single-person filtering |
| - ⏱️ clip segmentation into 5–10 second windows |
|
|
| This improves data quality for the music-driven dance generation task. |
|
|
| --- |
|
|
| ## 🎯 Intended Usage |
|
|
| This dataset is intended for research on: |
|
|
| - **music-driven dance video generation** |
| - **music-driven 3D dance generation** |
| - **pose-driven / motion-driven human animation** |
| - **motion and appearance evaluation** |
|
|
| --- |
|
|
| ## ⚠️ Notes |
|
|
| - This dataset is released for **research purposes only**. |
| - Please ensure your use complies with the corresponding license and platform policies. |
| - Some samples may originate from internet videos and are curated for academic research. |
|
|
| --- |
|
|
| ## 📚 Citation |
|
|
| If you find this dataset useful, please cite: |
|
|
| ```bibtex |
| @article{yang2026macedance, |
| title={MACE-Dance: Motion-Appearance Cascaded Experts for Music-Driven Dance Video Generation}, |
| author={Yang, Kaixing and Zhu, Jiashu and Tang, Xulong and Peng, Ziqiao and Zhang, Xiangyue and Wang, Puwei and Wu, Jiahong and Chu, Xiangxiang and Liu, Hongyan and He, Jun}, |
| journal={ACM Transactions on Graphics (SIGGRAPH 2026)}, |
| year={2026} |
| } |
| ``` |
