BoDiffusion

This is the official implementation of the paper: BoDiffusion: Diffusing Sparse Observations for Full-Body Human Motion Synthesis.

Paper

BoDiffusion: Diffusing Sparse Observations for Full-Body Human Motion Synthesis
Angela Castillo ^1*, María Escobar ^1*, Guillaume Jeanneret ², Albert Pumarola ³, Pablo Arbeláez¹, Ali Thabet ³, Artsiom Sanakoyeu ^3
*Equal contribution.
¹ Center for Research and Formation in Artificial Intelligence (CinfonIA), Universidad de Los Andes.
² University of Caen Normandie, ENSICAEN, CNRS, France.
³ Meta AI.

Dependencies and Installation

• Python >= 3.7 (Recommend to use Anaconda)
• PyTorch == 1.13.0 TorchVision == 0.15.2
• NVIDIA GPU + CUDA v10.1.243

1. Clone repo

   git clone https://github.com/BCV-Uniandes/BoDiffusion
   

2. Install dependent packages

   cd BoDiffusion
   conda env create -f env.yaml
   

Dataset Preparation

• Please refer to this repo for details about the dataset organization and split.

Train

• Training command:

  python train.py
  

• Pre-trained SR model: Find the pre-trained SR model at Drive.

Citations

If BoDiffusion helps your research, please consider citing us.

@inproceedings{castillo2023bodiffusion,
  title={BoDiffusion: Diffusing Sparse Observations for Full-Body Human Motion Synthesis},
  author={Castillo, Angela and Escobar, Maria and Jeannerete, Guillaume and Pumarola, Albert and Arbel{\'a}ez, Pablo and Thabet, Ali and Sanakoyeu, Artsiom},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  year={2023}
}

Find other resources in our webpage.

License and Acknowledgement

This project borrows heavily from Guided Diffusion, we thank the authors for their contributions to the community.

Contact

If you have any question, please email a.castillo13@uniandes.edu.co.