Neural Networks with Keras Cookbook

Getting ready

The strategy we'll adopt to generate images is as follows:

1. Generate a synthetic image using the generator network, which in the initial step is a noisy image that is generated by reshaping a set of noise values to the shape of our images.

2. Concatenate the generated image with the original set of images where the discriminator predicts whether each of the images is a generated image or an original image—this ensures that the discriminator is trained:
   • Note that the weights of a discriminator network are trained in this iteration
   • The loss of a discriminator network is the binary cross-entropy of the prediction and actual values of an image
   • The output value of the generated image will be fake (0) and the values of the original ...