Training Deep Convolutional Generative Adversarial Networks using tf.data.Iterator
MNIST images generated after 15000 global steps.
But DCGAN generates random images other than what we want.
Next time, I'll build a Conditional Generated Adversarial Networks that can produce the images we want.