Using Generative AI for Reconstructing Cultural Artifacts: Examples Using Roman Coins

Abstract

Generative AI, propelled by innovations like ChatGPT, has gained widespread recognition. In the realm of archeology, there exists significant potential for generative AI, particularly in reconstructing the appearance of cultural artifacts through the introduction of 2D or 3D renderings derived from damaged or degraded objects. In this study, we showcase and evaluate the practical application of Generative Adversarial Networks (GANs), harnessing the power of deep learning, for 2D image reconstruction of ancient Roman coins, aimed at aiding their improved visualization. Roman coins are chosen as our focal point due to their relative abundance and accessibility through online repositories and datasets. Our results demonstrate improved ability to enhance damaged or degraded coins, rendering them more similar to their better-preserved counterparts. In some instances, generated coins are virtually indistinguishable from the originals. The contribution of this work showcases the potential of GANs in assisting cultural heritage specialists and archeologists in recreating the appearance of damaged objects, thereby facilitating improved visualization of coins that are not well preserved. However, we also discuss when there might be limitations to using GANs in reconstructions. Although this work is tailored for ancient coins, GANs hold promise in application for other artifacts provided sufficient training data are available. We discuss how GANs can be applied and improve the appearance of artifact reconstructions, where we also provide relevant data used in this research.

Keywords

• generative
• artificial intelligence
• adversarial network
• coins
• roman
• artifacts
• reconstruction