Skip to content

Latest commit

 

History

History
19 lines (11 loc) · 1.07 KB

README.md

File metadata and controls

19 lines (11 loc) · 1.07 KB

Van-Gogh-me

This repository holds the source code for building and training a GAN to generate celeb portraits painted in, artist, Van Gogh's painting style. We experimented with GANs in this project.

Directory Structure

The repository has 3 sub directories, namely,

  • models This directory, in turn, contains the sub-directory:

    • Scripts: This directory contains scripts to build parts of the GAN such as the generator (generator.py) and discriminator (discriminator.py) networks in addition to their combination i.e. the final CycleGAN architecture (CycleGAN.py). This directory also contains a module (losses.py) collecting all loss functions used in training the CycleGAN model.

  • notebooks This directory contains notebook which can be used to experiment with the constructed model.

  • preprocessing This directory contains the script for collecting the painting images and celeb images in 2 different tf.data.Datasets objects. These objects when zipped together can be passed to the model.fit method, as done in models/Scripts/train.py.