/gan-tests

A collection of Generative Adversarial Networks (GANs)

Primary LanguagePython

GAN Tests

Implementations

vanillaGAN.py: Code taken from this blog post. It trains a GAN on MNIST data.

  • The generator takes a 100-dimensional noise vector and returns a 786-dimensional vector. This return vector is the same size as an MNIST image (28x28). In a way, it learns a mapping between the prior space (100-dim vector) and the MNIST data.
  • The discriminator takes in an MNIST image and returns a scalar which represents the probability that the image passed in is a real MNIST image.
  • Samples from the generator are generated every 1000 batches run over the MNIST data. The resulting samples can be viewed in the out folder.
  • Analysis: After ~300,000 training iterations (~1 hr), the generator creates passable MNIST digits.