Image-driven generative cell modelling with adversarial autoencoders
Installing on linux is recommended.
Running on docker is recommended, though not required.
- install torch on docker / nvidia-docker as in e.g. this guide: https://github.com/gregjohnso/dl-docker
- download the training images:
aws s3 cp s3://aics.integrated.cell.arxiv.paper.data . --recursive --no-sign-request
After you clone this repository, you will need to edit the mount points for the images in run_docker.sh to point to where you saved them.
Once those locations are properly set, you can start the docker image with
bash run_docker.sh
Once you're in the docker container, you can train the model with
bash train_model_2D.sh
This will take a while, probably about 12-18 hours.
Example outputs of this model can be viewed at http://www.allencell.org
If you find this code useful in your research, please consider citing the following paper:
@article{johnson2017generative,
title={Generative Modeling with Conditional Autoencoders: Building an Integrated Cell},
author={Gregory R. Johnson, Rory M. Donovan-Maiye, Mary M. Maleckar},
journal={arXiv preprint arXiv:1705.00092},
year={2017},
url={https://arxiv.org/abs/1705.00092}
}
Gregory Johnson E-mail: gregj@alleninstitute.org