The aims of this project was to colorize grayscale portrets.
The data used in this project is the CelebA dataset. You can download the zip file from this site: https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html. In the file 'data_preprocessing.ipynb,' you can find ways to process this data and generate datasets in LAB and RGB color spaces.
In this project, three architectures of CNN were used:
- Sequential CNN ('beta-model.ipynb')
- An Autoencoder ('autoencoder.ipynb')
- CNN and Inception-ResNet-v2 (Deep Koalarization) ('inception-model.ipynb') About the third architecture, you can read more here: https://arxiv.org/abs/1712.03400
The best result was received with the CNN + Inception-ResNet-v2 model. You can find the code in 'best_model.ipynb'.
