Skip to content

maxcaseylin/album_gen

Repository files navigation

album_gen

Code for training a DCGAN to generate pictures of album art, written in Tensorflow, along with:

  • scripts for preprocessing and cleaning images to fit in the neural network
  • a script to combine the output images during training into a GIF

Demos

Result after 1.5 hours(300 Epochs) of training, with 1.2k images and batch size 100

output_1k

Result after about 5 hours(186 Epochs) of training, with 10k images and batch size 128

output_10k

Result on celeba dataset after 1 hour (350 epochs of training), with 1k images and 128 batch size

output_celeb_1k

Result on celeba dataset after 5 hours (350 epochs of training), with 4.7k images and 128 batch size

output_celeb_4k

Due to hardware limitations I haven't been able to train the network as much as I would have liked, so in the future if I have better hardware I'll probably get better results

Prerequisites

  • Tensorflow

  • matplotlib

  • imageio(if using the GIF script)

Usage

Personal usage of this repository is NOT recommended as data processing and cleaning are all separate scripts.

About

GAN for generating album cover art

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published