KungFu-Fiesta (KFF) Dataset

The first martial arts combat video dataset for personalized two-person fighting video generation.

This dataset was introduced in the paper MagicFight: Personalized Martial Arts Combat Video Generation (ACM MM 2024).

📖 Overview

KungFu-Fiesta is a specialized dataset designed for the novel task of Personalized Martial Arts Combat Video Generation. Unlike existing single-person dance video datasets, KFF focuses on two-person interactions in martial arts combat scenarios, addressing challenges such as:

• Identity confusion between two fighters
• Anomalous limbs during complex movements
• Action mismatches in combat sequences

The dataset was created using Unity, an advanced game physics engine, to generate highly realistic 3D character models and action animations in simulated environments.

📊 Dataset Statistics

⚠️ Note: This is a partial release of the KungFu-Fiesta dataset. The full dataset contains 500+ videos with 12 scenes, 12 action types, and 12 character types as described in the paper. Some data remains private and is not publicly available.

Attribute	Combat Videos	DWPose Sequences
Total Count	293	293
Resolution	1728 × 1080	832 × 512
Frame Rate	60 FPS	60 FPS
Duration	2-13.5s	2-13.5s

📁 Dataset Structure

KungFu-Fiesta/
├── README.md
├── data/
│   ├── videos/
│   │   ├── metadata.csv
│   │   ├── 001.mp4
│   │   └── ...
│   └── dwpose/
│       ├── 001.mp4
│       └── ...

Each Sample Contains

• Combat video: Two-person martial arts fighting video
• DWPose sequence: Corresponding pose skeleton video (body, hands, and foot keypoints)

🚀 Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("MingfuYAN/KungFu-Fiesta")

# Access metadata
for item in dataset['train']:
    video_path = item['file_name']
    dwpose_path = item['dwpose_file_name']

Metadata Fields

Field	Description
file_name	Relative path to the video file
dwpose_file_name	Relative path to the DWPose skeleton video

📜 License

This dataset is released under CC BY-NC 4.0 (Creative Commons Attribution-NonCommercial 4.0 International).

📝 Citation

If you use this dataset in your research, please cite:

@inproceedings{huang2024magicfight,
  title={Magicfight: Personalized martial arts combat video generation},
  author={Huang, Jiancheng and Yan, Mingfu and Chen, Songyan and Huang, Yi and Chen, Shifeng},
  booktitle={Proceedings of the 32nd ACM International Conference on Multimedia},
  pages={10833--10842},
  year={2024}
}

📧 Contact

For questions about the dataset, please contact:

• Jiancheng Huang: jc.huang@siat.ac.cn
• Mingfu Yan: mfyan99@gmail.com