GANBan

Correcting and coloring old and damaged images using a CycleGAN

Converting Images

python convert.py input_file

This will fix a damaged image and and will colorize grayscale images.

This requires the weights ganban.pth to be in the same directory. This can be downloaded from here.

Additional arguments are as follows:

usage: convert.py [-h] [-i] [-m] [-r] input_file [input_file ...]

Converts an old damaged image into a new colored image. Outputs a file(s) with the name input_file.<step>.png

positional arguments:
  input_file          input file(s)

options:
  -h, --help          show this help message and exit
  -i, --intermediate  Output (save) the intermediate states
  -m, --multi-domain  Use the multiple domain/model approach (damaged<>fixed<>color)
  -r, --reverse       Reverses the process (corrected image --> old image)

Dataset

You can download the dataset used here. It contains ~900 unique images trained on for 3 different models.

Structure

data

Dataset location and directory to save models in.

config.py

Contains the hyperparameters.

cyclegan.py

Contains the generator, discriminator, and initialization weights.

datasetup.py

Contains the class (Dataset) that interacts with the data

engine.py

Setup for the train and test steps

train.py

Entry point to begin training the model. To start, execute python train.py.