中文文档 ｜ Tutorial Video (coming soon)

❓ What is UniMoCap?

UniMoCap is a community implementation to unify the text-motion mocap datasets. In this repository, we unify the AMASS-based text-motion datasets (HumanML3D, BABEL, and KIT-ML). We support to process the AMASS data to both :

• body-only H3D-format (263-dim, 24 joints)
• whole-body SMPL-X-format (322-dim SMPL-X parameters).

We believe this repository will be useful for training models on larger mocap text-motion data. We will support more T-M mocap datasets in near feature.

We make the data processing as simple as possible. For those who are not familiar with the datasets, we will provide a video tutorial to tell you how to do it in the following weeks. This is a community implementation to support text-motion datasets. For the Chinese community, we provide a Chinese document (中文文档) for users.

🏃🏼 TODO List

• Support SMPL-X motion representation (including position, velocity, and rotations for body and hands) calculation (expected in 1-2 week).
• Support seg only and seq only BABEL unifier.
• Provide a tutorial video.
• Support more language documents.
• Support more T-M datasets (e.g.: FLAG3D, STDM). Welcome to support your own dataset here!
• Provide trained models based on UniMoCap.

🛠️ Installation

pip install -r requirements.txt

🚀 How to use?

1. Data Preparing

Download SMPL+H and DMPLs.

Download SMPL+H mode from SMPL+H (choose Extended SMPL+H model used in AMASS project), DMPL model from DMPL (choose DMPLs compatible with SMPL), and SMPL-X model from SMPL-X. Then place all the models under ./body_model/. The ./body_model/ folder tree should be:

./body_models
├── dmpls
│   ├── female
│   │   └── model.npz
│   ├── male
│   │   └── model.npz
│   └── neutral
│       └── model.npz
├── smplh
│   ├── female
│   │   └── model.npz
│   ├── info.txt
│   ├── male
│   │   └── model.npz
│   └── neutral
│       └── model.npz
├── smplx
│   ├── female
│   │   ├── model.npz
│   │   └── model.pkl
│   ├── male
│   │   ├── model.npz
│   │   └── model.pkl
│   └── neutral
│       ├── model.npz
└───────└── model.pkl

Download AMASS motions.

• Download AMASS motions.
• If you are using the SMPL (in HumanML3D, BABEL, and KIT-ML), please download the AMASS data with SMPL-H G into ./datasets/amass_data/.
• If you are using the SMPL-X (in Motion-X), please download the AMASS data with SMPL-X G. If you use the SMPL-X data, please save them at ./datasets/amass_data-x/.

The datasets/amass_data/ folder tree should be:

./datasets/amass_data/
├── ACCAD
├── BioMotionLab_NTroje
├── BMLhandball
├── BMLmovi
├── CMU
├── DanceDB
├── DFaust_67
├── EKUT
├── Eyes_Japan_Dataset
├── GRAB
├── HUMAN4D
├── HumanEva
├── KIT
├── MPI_HDM05
├── MPI_Limits
├── MPI_mosh
├── SFU
├── SOMA
├── SSM_synced
├── TCD_handMocap
├── TotalCapture
└── Transitions_mocap

HumanML3D Dataset

Clone the HumanML3D repo to datasets/HumanML3D/, unzip the humanact12.zip to datasets/pose_data and unzip the texts.zip file.

mkdir datasets
cd datasets
git clone https://github.com/EricGuo5513/HumanML3D/tree/main
mkdir pose_data
unzip HumanML3D/pose_data/humanact12.zip -d pose_data/
mv pose_data/humanact12/humanact12/*.npy pose_data/humanact12/
rm -rf pose_data/humanact12/humanact12
cd HumanML3D/HumanML3D
unzip texts.zip
cd ../../..

KIT-ML Dataset

Download KIT-ML motions, and unzip in the folder datasets/kit-mocap/.

BABEL Dataset

Download the BABEL annotations from TEACH into datasets/babel-teach/.

2. Generate mapping files and text files

In this step, we will get mapping files (.csv) and text files (./{dataset}_new_text).

HumanML3D Dataset

Due to the HumanML3D dataset is under the MIT License, I have preprocessed the .json (./outputs-json/humanml3d.json) file and .csv file (h3d_h3dformat.csv). Besides, the .csv file can be generated by the following command.

python h3d_to_h3d.py

BABEL Dataset

We provide the code to generate unified BABEL annotation. Both .json (./outputs-json/babel{_mode}.json) file and .csv file (babel{mode}_h3dformat.csv) are generated. You can generate related files with the following command. The .json file is only an intermediate generated file and will not be used in subsequent processing.

For BABEL, babel_seg and babel_seq denote the segmentation level and whole-sequence level annotation respectively. The babel denotes the both level annotation.

python babel.py

KIT-ML Dataset

We provide the code to generate unified KIT-ML annotation. Both .json (./outputs-json/kitml.json) file and .csv file (kitml_h3dformat.csv) are generated. You can generate related files with the following command. The .json file is only an intermediate generated file and will not be used in subsequent processing.

python kitml.py

3. Extract and Process Data

In this step, we follow the method in HumanML3D to extract motion feature.

Now, you are at the root position of ./UniMoCap. To generated body-only motion feature, we create a folder and copy some tools provided by HumanML3D first:

mkdir body-only-unimocap
cp -r ./datasets/HumanML3D/common ./
cp -r ./datasets/HumanML3D/human_body_prior ./
cp -r ./datasets/HumanML3D/paramUtil.py ./

If you would like to get body-only motions, please refer to Body-only data preprocessing.

If you would like to get whole-body motions, please refer to Whole-body motion processing.

💖 Opening for Community Contributions

We sincerely wish the community to support more text-motion mocap datasets.

🌹 Acknowledgement

Our code is modified on the basis of TMR, AMASS-Annotation-Unifier, and HumanML3D, thanks to all contributors!

Star History

🤝🏼 Citation

If you use this repository for research, you need to cite:

@article{chen2023unimocap,
  title={UniMocap: Unifier for BABEL, HumanML3D, and KIT},
  author={Chen, Ling-Hao and UniMocap, Contributors},
  journal={https://github.com/LinghaoChan/UniMoCap},
  year={2023}
}

As some components of UniMoCap are borrowed from AMASS-Annotation-Unifier and HumanML3D. You need to cite them accordingly.

@inproceedings{petrovich23tmr,
    title     = {{TMR}: Text-to-Motion Retrieval Using Contrastive {3D} Human Motion Synthesis},
    author    = {Petrovich, Mathis and Black, Michael J. and Varol, G{\"u}l},
    booktitle = {International Conference on Computer Vision ({ICCV})},
    year      = {2023}
}

@InProceedings{Guo_2022_CVPR,
    author    = {Guo, Chuan and Zou, Shihao and Zuo, Xinxin and Wang, Sen and Ji, Wei and Li, Xingyu and Cheng, Li},
    title     = {Generating Diverse and Natural 3D Human Motions From Text},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2022},
    pages     = {5152-5161}
}

If you use the dataset, you need to cite subset KIT-ML and AMASS.

@article{Plappert2016,
    author = {Matthias Plappert and Christian Mandery and Tamim Asfour},
    title = {The {KIT} Motion-Language Dataset},
    journal = {Big Data}
    publisher = {Mary Ann Liebert Inc},
    year = {2016},
    month = {dec},
    volume = {4},
    number = {4},
    pages = {236--252}
}

@conference{AMASS2019,
  title = {AMASS: Archive of Motion Capture as Surface Shapes},
  author = {Mahmood, Naureen and Ghorbani, Nima and Troje, Nikolaus F. and Pons-Moll, Gerard and Black, Michael J.},
  booktitle = {International Conference on Computer Vision},
  pages = {5442--5451},
  month = oct,
  year = {2019},
  month_numeric = {10}
}

If you use the Motion-X dataset, please cite it accordingly.

@article{lin2023motionx,
  title={Motion-X: A Large-scale 3D Expressive Whole-body Human Motion Dataset},
  author={Lin, Jing and Zeng, Ailing and Lu, Shunlin and Cai, Yuanhao and Zhang, Ruimao and Wang, Haoqian and Zhang, Lei},
  journal={Advances in Neural Information Processing Systems},
  year={2023}
}

If you have any question, please contact Ling-Hao CHEN (thu [DOT] lhchen [AT] gmail [DOT] com).