/GAN-tests

Various experiments with GAN architectures

Primary LanguageJupyter Notebook

GAN Experiments

These are experiments of trying to train GANS from scratch.

Each notebook has all the code needed to train GAN. There are 4 notebooks that differ in techniques used for training.

Below are results.

Vanilla

Vanilla

Minibatch Discrimination

Minibatch Discrimination

Augmenting images

Augments

Multiscale gradient + pixelwise norm + minibatch std + equalized learning + relativistic hinge loss

Many

DCGAN

DCGAN

DCGAN + Spectral Norm

Spectral Norm