/colorizer_GAN

Colorize Images with a Generative Adversarial Network (GAN)

Primary LanguageJupyter Notebook

colorizer_GAN w/ Lightning

This repository is primarily meant to showcase the features of Pytorch Lightning and how it can streamline three different training tasks. The training tasks in order are:

  1. Self-Supervized Training of the Generator Backbone (Resnet 18)
  2. Supervized Training of the Generator
  3. GAN Training

GAN Colorizer from Scratch (No Pretrained weights used!)

You can train your own GAN Colorizer, or use the given weights to colorize your own images. The model that produced the images above was trained for a total of 70 epochs all within 3 hours. Most of the GAN architecture code and generator training implementation was adapted from Moein's GAN Colorizer training. I also took most of the Self-Supervized dataset code from Ian's Implementation of Self-Supervized Learning.