Skip to content

lyshuga/PokemonDCGAN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 

Repository files navigation

PokemonDCGAN

The project is assigned to the topic of Generating Pokemon images using Deep Convolutional GAN. Convolution layers allow to identify and generate precise representation fo the image. Though in DCGAN it is the main obstacle as far as training two networks (Generator and Discriminator) is a pretty sensitive process.

Eventually I came out that Generator should be pretty simple, meanwhile, Discriminator should be complex in order to prevent constant Discriminator winning.

That is why Discriminator is:

alt text

And Generator is much simplier:

alt text

During the training with augmentation, assigning different values for parameters and changing architecture, I have got such results:

alt text

Though not as impressive as expected. The future plan is to make the generator more complex while trying to keep the discriminator the same.

About

Creating images of Pokemon using DCGAN

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published