Robust Dancer: Long-term 3D Dance Synthesis Using Unpaired Data

Get started

dependencies required

• python >=3.7
• pytorch
• smplx
• aistplusplus_api

Get the checkpoint

get trained checkpoint from: https://drive.google.com/drive/folders/1Z-kghUqJLEbjgzg2soHsmnWoM62r5EzL?usp=sharing

Run the code

1. get the data from aistpp website: https://google.github.io/aistplusplus_dataset/

2. data preprocess

python data_preprocess/extract_aist_feature.py  # necessary for test
python data_preprocess/motion_data_process.py  #necessary for train

3. run training

python train.py

4. run inference and test

python inference.py
python test_scripts/calculate_beat_scores.py
python test_scripts/calculate_fid_scores.py
python test_scripts/train_style_classi.py
python test_scripts/style_classi_new.py

Citation

@article{Feng2023RobustDL,
  title={Robust Dancer: Long-term 3D Dance Synthesis Using Unpaired Data},
  author={Bin Feng and Tenglong Ao and Zequn Liu and Wei Ju and Libin Liu and Ming Zhang},
  journal={ArXiv},
  year={2023},
  volume={abs/2303.16856}
}