Image Colourization using Conditional Generative Adversarial Networks

Image Colourization

• Final project for the Machine Learning Master's course at University of Groningen
• The focus is on Colourization task using Conditional Generative Adversarial Networks
• Using pre-trained ResNet and UNet
• Some code is borrowed from one of the author's repo

Instructions to run the code

• For running the train script use. Use --help to list all the commandline options.

python3 src/train.py --help

• After training, use the following script to compute evaluation metrics on the test set. Use --help to list all the commandline options.

python3 src/compute_test_metrics.py --help

• After training, use the following script to colourize the images using the Generator network. Use --help to list all the commandline options.

python3 src/generate_results.py --help

• Use notebook src/plot_train_losses.ipynb to visualize training losses
• Use notebook src/visualize_results.ipynb to visualize the generated results from Generator and compare it with the original images.

Weights file

• The weights file is available

Dependencies

• The dependencies are available in the requirements.txt

Reference

• Image-to-Image Translation with Conditional Adversarial Networks