Synthetic COVID-19 CXR Dataset

Synthesis of COVID-19 chest X-rays using unpaired image-to-image translation

This work builds the first-of-its-kind open dataset of synthetic COVID-19 chest X-ray images of high fidelity using an unsupervised domain adaptation approach by leveraging class conditioning and adversarial training, and shows considerable performance improvements on CO VID-19 detection using various deep learning architectures when employing synthetic images as additional training set.