I am writing a custom training loop for GANs and at some point I want to both evaluate my model metrics and use its output (images), in this case to draw the images in tensorboard.
I wanted to know if there was a way to efficiently do evaluate_generator and predict_generator at the same time.
I saw this Github issue, but it has been marked as stale without intervention from keras developpers. Since not a lot of people seem to have this issue, I guess there is a way to do it efficiently, but I haven't figured out how.