AI Projects

Mosaic.AI at GTC'23 

This project was accepted at NVIDIA GTC'23. Please check this link after registration to see more. 

Generative Adversarial Neural Networks For Roman Mosaic Restoration in Africa [PS51223]

In order to make a substantial jump forward, we need to take a step back and learn from our ancestors. Carefully studying ancient Roman mosaics is one of way of doing that. Most of the mosaics in the North African region of Algeria are rapidly degrading for various reasons. The purpose of this project, which is an artificial Intelligence tool, is to restore mosaics to their original state. We experimented with two different methods that will be discussed in the methodology section.

Primary Topic: Computer Vision

Industry: Architecture / Engineering / Construction


The Mosaic Restoration is the result of a long sequence of study phases, treatments and interventions with the contribution of technology.

The restoration process must be focused not only on the surface, but on the structure, the materials that constitute the mosaic, etc. Therefore it is essential to understand the causes of deterioration or the degradation and the problems that have caused enough damage to warrant restoration. The integrity of work does not depend on its age, but on other factors such as materials, the technique used, and the exposure conditions (environment, indoor or outdoor) as well as its use (wall hangings, floors, sculptures, urban decorations, architecture, etc.). There is a need for professionals capable to restore the functionality of those mosaics that are generally private, and not subject to liability.