Fast Colorization of Grayscale Images by Convolutional Neural Network.
This model uses interconversion between RGB and LAB color spaces to find traits that link each grayscale value to corresponding colour values. Currently, it has only been trained on a dataset of 700 human faces with accuracy ~ 60%.
- Improve the network to provide colorization for not only for faces but also objects.
- Train the network with a larger dataset to obtain better results and improve accuracy.
You are welcome to open pull requests :)