As an effort to recreate the Paris of 1970 by showcasing the colorized versions of the photos in the amateur contest “C’était Paris en 1970,” we discovered an issue with the current models for image colorization: they lacked the ability to take into consideration metadata of the image, such as time and location, that could vary its probable colors. Therefore, we created a metric for evaluating the performance of colorization algorithms for distinct places around the world, and we propose an algorithm that can build upon the results of the model to get the image to resemble probable colors in the given location. After assessing our results by the same metric used for the evaluation of the generic algorithm, we found improvement in the colorization.

Through this SuperUROP, I have the intent to gain high-level research experience while applying my knowledge of machine learning to image analysis and classification. I am also interested in analyzing the subtle displays of political biases in photography. I hope this project yields meaningful results that can be included in a paper by the end of the SuperUROP.