BioMime

Pytorch implementation and pretrained models for BioMime. For details, see Human Biophysics as Network Weights: Conditional Generative Models for Dynamic Simulation

Requirements

• Operating System: Linux.
• Python 3.7.11
• PyTorch >= 1.6
• torchvision >= 0.8.0
• CUDA toolkit 10.1 or newer, cuDNN 7.6.3 or newer.

Conda environment

environment.yml contains all the dependencies required to run BioMime. Create the new environment by:

conda env create --file environment.yml

Data for training

Please contact neurodec for the dataset.

Pretrained models

Download model.pth and put them under ckp/.

Quick Start

Train

When you have your data ready, please follow the instructions below to train your own BioMime:

1. Edit utils/data.py to specify the path for dataset.
2. Configure the models and setting up in config/config.yaml.
3. Run the training script by:

python scripts/train.py --exp=test

Define your own experiment id by changing the argument --exp.

Test

The checkpoints at snapshot epochs will be saved in res/exp/. You can test the model by:

python scripts/test.py --ckp_pth=./ckp/linear_anneal.pth --num_sample=32 --plot=1

Generate

You can generate your own MUAPs by sampling from the standard Normal Distribution:

python scripts/generate.py --cfg config.yaml --mode sample --model_pth ./ckp/model_linear.pth --res_path ./res

Or by morphing the existing MUAPs:

python generate.py --cfg config.yaml --mode morph --model_pth ./ckp/model_linear.pth --res_path ./res

Make sure you have the file containing MUAPs in the format of [num, nrow, ncol, ntime] and set the argument --data_path. Examples of MUAP files will be provided in the future.

We also allow users to generate dynamic MUAPs during a realistic forearm movement defined by a musculoskeletal model. This new function will be available soon.

Install BioMime as a python package

pip install git+https://github.com/shihan-ma/BioMime.git

Uninstall BioMime before updating it.

pip uninstall BioMime

Licenses

This repository is released under the GNU General Public License v3.0.

Citation

@article{ma2022human,
  title={Human Biophysics as Network Weights: Conditional Generative Models for Ultra-fast Simulation},
  author={Ma, Shihan and Clarke, Alexander Kenneth and Maksymenko, Kostiantyn and Deslauriers-Gauthier, Samuel and Sheng, Xinjun and Zhu, Xiangyang and Farina, Dario},
  journal={arXiv preprint arXiv:2211.01856},
  year={2022}
}