Revist

So you mean the autoencoder will work fine even if I supply some images from different resolution / distribution (e.g. it has not seen images with cats and I supply...

I see, then the question is how to use one tar file for multiple models? These tars tend to be big.

I agree, I believe the code ought to be: batch_imgs = np.concatenate(imgs, gen_imgs) batch_labels = np.concatenate(valid, fake) d_loss = self.discriminator.train_on_batch(batch_imgs, batch_labels) d_loss *= 1/batch_size (not sure where 0.5 came from...