Generative Adversarial Networks - GANs Intro

GAN Input Code Demo - Reshaping and Projecting Data

When speaking about the input to a GAN generator, we usually have stated that the input is a batch of noise vectors with values randomly sampled from a standard normal distribution. We can think of a batch in this way as having the shape (batch_size, 100).

As we saw when we implemented a DCGAN in PyTorch and in TensorFlow, however, the input wasn't actually a batch of 1-dimensional noise vectors of length 100. Instead, the input was a batch having shape (batch_size, 100, 1, 1) for PyTorch, or (batch_size, 1, 1, 100) for TensorFlow.