How can I obtain the subset of a Keras testing dataset for which a model makes correct predictions?

Question

I am working on a system that compares and demonstrates the effects of adversarial attacks on images from different datasets and how they impact the performance of Keras/Ten