This repository hosts dataset for our ICML 2021 Workshop on Computational Biology (WCB) paper. A more detailed version titled Synthesis of COVID-19 Chest X-rays using Unpaired Image-to-Image Translation is published in the Journal of Social Network Analysis and Mining (SNAM), Special Issue on Tackling COVID-19 Infodemic.
The dataset consists of 21,295 synthetic COVID-19 chest X-ray images generated using this algorithm. Dataset is available at this link.
Figure 1. Illustration of the data generation process based on unpaired image-to-image translation. Chest X-ray images are translated from Non-COVID-19 (i.e. Normal or Pneumonia) to COVID-19 and then back to Non-COVID-19 via cycle-consistency
Here's a video of the learning in progress. Top row (Normal CXR, Translated COVID-19 CXR, Reconstructed Normal CXR), bottom row (COVID-19 CXR, Translated Normal CXR, Reconstructed COVID-19 CXR).
If you use this dataset in your scientific work, please cite the following:
@article{zunair2021synthesis,
title={Synthesis of {COVID}-19 chest {X}-rays using unpaired image-to-image translation},
author={Zunair, Hasib and Hamza, A Ben},
journal={Social Network Analysis and Mining},
volume={11},
number={1},
pages={1--12},
year={2021},
publisher={Springer}
}
The synthetic dataset was generated using https://github.com/hasibzunair/adversarial-lesions.
