Skip to content

dremovd/yadcgan-keras

Repository files navigation

yadcgan-keras

Yet Another DCGAN for keras

The aim of DCGAN model is to be able to produce images that would be similar to the train set.

Dataset consists of ~87000 RGB images of size 48x48 from old games. Mainly images are of characters and objects. Images are at original scale. Dataset was extracted from http://spritedatabase.net/ and published with explicit permit of Paul, Sprite Database owner

Requirements:

  • python 2.7
  • keras with theano
  • opencv3, for saving files
  • joblib, to load image dump of big size
  • matplotlib, for debug output
  • GPU card for reasonable training speed

Examples:

After ~5 hours of training After ~16 hours of training

After ~30 hours of training After ~40 hours of training

About

Yet Another DCGAN for keras

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published