Generalization of COVID-19 models is a very serious challenge. I'm trying to find a model (ct scans or x-rays) that was trained and evaluated on multiple datasets. It would be particular interesting to look at the model that was evaluated on an entirely different dataset (not in the training set.