Hands-On Transfer Learning with Python

Problem statement revisited

If we follow the most widely used color model, RGB, it turns out that training a model to map an input monochrome image to a color one would be quite a daunting task.

Researchers in the deep learning space have been quite creative when it comes to approaching and formulating a problem. In the case of image colorization, researchers have cleverly looked into ways of utilizing different input to achieve realistic color hallucinations of grayscale images.

In the initial attempts, different variants of color-guided input, in the form of reference images and color scribbles, were used to a great effect; see Welsh and co-authors as well as Levin and co-authors.

Recent works have focused on automating the entire process ...