Small project testing colorization of black-white/greyscaled images using some neural networks.
Takes the simple bw version of the images as base and tries to generate the original gt.
Train_with_demo.ipynb contains the main code, the other notebook were created for testing purposes.
Currently the model used is an UNET model using effnet-b2. Other architectures might perform better, especially GANs (as other projects have shown).
Here's a small example gif how the images of the test set can change per epoch.
There are some obvious problem atm and various possible improvements that can be implemented. I will test them if I find time and motivation.