[3DV2021] Dynamic Multi-Person Mesh Recovery From Uncalibrated Multi-View Cameras (DMMR)

The code for 3DV 2021 paper "Dynamic Multi-Person Mesh Recovery From Uncalibrated Multi-View Cameras"
Buzhen Huang, Yuan Shu, Tianshu Zhang, Yangang Wang
[Paper] [Video]

Dependencies

Windows or Linux, Python3.7

conda install pytorch torchvision torchaudio cudatoolkit=11.3 -c pytorch
pip install -r requirements.txt

Getting Started

Step1:
Download the official SMPL model from SMPLify website and put it in models/smpl. (see models/smpl/readme.txt)

Step2:
Download the test data and trained motion prior from Google Drive or Baidu Netdisk (extraction code [jomn]) and put them in data.

Step3:
Run

python main.py --config cfg_files/fit_smpl.yaml

You can visualize the motions and cameras in optimization with the command:

python main.py --config cfg_files/fit_smpl.yaml --visualize true

The code can also be used for motion capture with known cameras:

python main.py --config cfg_files/fit_smpl.yaml --opt_cam false

Results

The fitted results will be saved in output.
You can visualize the estimated extrinsic camera parameters by running:

python viz_cameras.py

Citation

If you find this code useful for your research, please consider citing the paper.

@article{huang2023simultaneously,
  title={Simultaneously Recovering Multi-Person Meshes and Multi-View Cameras with Human Semantics},
  author={Huang, Buzhen and Ju, Jingyi and Shu, Yuan and Wang, Yangang},
  journal={IEEE Transactions on Circuits and Systems for Video Technology},
  year={2023},
  publisher={IEEE}
}
@inproceedings{huang2021dynamic,
      title={Dynamic Multi-Person Mesh Recovery From Uncalibrated Multi-View Cameras}, 
      author={Buzhen Huang and Yuan Shu and Tianshu Zhang and Yangang Wang},
      year={2021},
      booktitle={3DV},
}

Acknowledgments

Some of the code are based on the following works. We gratefully appreciate the impact it has on our work.
SMPLify-x
SPIN
EasyMocap
MvSMPLfitting