Fig. 6.

Confidence scores for the VGG-16 model on unseen test set of COVID-19 for the two binary classification tasks of Normal vs. COVID-19 (left) and Pneumonia vs. COVID-19 (right) with synthetic data as additional training set. Left: adding 16,537 COVID-19 examples of ${\mathcal{G}}_{\mathit{NC}}$ to the original COVID-19 dataset. Right: adding 4758 COVID-19 examples of ${\mathcal{G}}_{\mathit{PC}}$ to the original COVID-19 dataset