Case Study in Scientific Computing

This was a group project for the Mathematical Modelling and Scientific Computing MSc in which we created a GUI for recolorising black and white images with limited colour information. The underlying algorithm relies on reproducing kernel hilbert spaces. We also investigated optimising the parameters of the model using bayesian optimisation.

Installation

Clone the repository:
git clone https://github.com/bennm37/image_colourisation.git
Change directory to the repository:
cd image_colourisation
Create a virtual environment:
python -m venv venv
NOTE python>=3.10 is required for match statements. Activate the virtual environment:
source venv/bin/activate
Install the requirements:
pip install -r requirements.txt
Good to go! Try running gui/main.py

Working Updates

GUI

• main.py seems to be largely functional as of 02-01.
• main.py needs to be ran from within the /gui/ folder at the moment and will only work on linux or mac unless the file paths within the script are changed to windows format. UPDATE 30-01: should work on windows, need to test.

Optimisation

Going to work on narrowing this down but so far we have:

• delta = 0.0200503376714422,
• sigma_1 = 96
• sigma_2 = 87.09553004603107,
• rho = 0.5644138192495579.

Images

Within trainingImages there are 4 subdirectories:

1. 256px square cartoon images
2. 256px square 'real life' images
3. 512px square cartoon images
4. 512px square 'real life' images