MACE-Dance

# 🎵 MACE-Dance Dataset **MACE-Dance** is a large-scale dataset for **music-driven dance video generation**, released with our SIGGRAPH 2026 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 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} } ``` ---