Image Colorization is one of the classical problems of Image to Image Translation wherein we input a grayscale image and output a colored image based on a coloring scheme.

With the recent advances in Deep Learning, a wide range of techniques have been proposed for automating the task of colorization of grayscale images. In our project, we study and compare the impact of various Deep Learning methods and techniques which have produced outstanding results in the problems of Image to Image translation. We study and develop some unique fusion techniques and understand the impact of different hyperparameters on the learning model.

Our project is not about increasing the accuracy for the task on some baseline but rather understanding the ramifications of changing different aspects of the learning model. The detailed problem statement can be found on the next page.

This project is undertaken by Anupraas Gautam and Harsh Rawat as a part of the Computer Vision course (CS 766) at the University of Wisconsin - Madison under the guidance of Professor Mohit Gupta.