MotionGPT

The official PyTorch implementation of the paper "MotionGPT: Human Motion Synthesis with Improved Diversity and Realism via GPT-3 Prompting".

Bibtex

If you find this code useful in your research, please cite:

@inproceedings{ribeiro2024motiongpt,
  title={MotionGPT: Human Motion Synthesis with Improved Diversity and Realism via GPT-3 Prompting},
  author={Ribeiro-Gomes, Jose and Cai, Tianhui and Milacski, Zolt{\'a}n A and Wu, Chen and Prakash, Aayush and Takagi, Shingo and Aubel, Amaury and Kim, Daeil and Bernardino, Alexandre and De La Torre, Fernando},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={5070--5080},
  year={2024}
}

NOTE: WIP

This code currently only has the instructions for inference. Training and data preparation will come shortly.

Getting started

This code was tested on Ubuntu 18.04 LTS and requires:

• Python 3.7
• conda3 or miniconda3
• CUDA capable GPU (tested on NVidia RTX A4000 16GB)

1. Setup environment

Install ffmpeg (if not already installed):

sudo apt update
sudo apt install ffmpeg

Setup conda env:

conda env create -f environment.yml
conda activate motiongpt
python -m spacy download en_core_web_sm
pip install git+https://github.com/openai/CLIP.git
pip install sentence_transformers 

Download dependencies:

bash prepare/download_smpl_files.sh
bash prepare/download_glove.sh
bash prepare/download_t2m_evaluators.sh

2. Download the pretrained models

Download the model(s) you wish to use, then unzip and place them in ./save/.

link

Motion Synthesis

Generate a single prompt

python -m sample.generate --model_path ./save/mini/model000600161.pt --text_prompt "greet a friend" --babel_prompt "hug"

You may also define:

• --device id.
• --seed to sample different prompts.
• --motion_length (text-to-motion only) in seconds (maximum is 9.8[sec]).
• --second_llm

Running those will get you:

• results.npy file with text prompts and xyz positions of the generated animation
• sample##_rep##.mp4 - a stick figure animation for each generated motion.

It will look something like this:

You can stop here, or render the SMPL mesh using the following script.

Render SMPL mesh

To create SMPL mesh per frame run:

python -m visualize.render_mesh --input_path /path/to/mp4/stick/figure/file

This script outputs:

• sample##_rep##_smpl_params.npy - SMPL parameters (thetas, root translations, vertices and faces)
• sample##_rep##_obj - Mesh per frame in .obj format.

Acknowledgments

This code is heavily adapted from:

• MDM

• HumanML3D

• MotionCLIP

• SMPL(-X)