Scaling Art Classification Models: Enhancing Binary Classifiers and Tackling the Challenge of AI-Generated Art

Abstract

This paper discusses the creation of an AI-based binary classification model that can efficiently classify artwork by artists, with two categories building on previous research in the area. The research topic here is to streamline current multi-classification models by using a binary classifier in assessing its performance compared to conventional multi-classification systems with multiple categories and artists simultaneously. Deep learning methods, specifically ResNet-101, were employed to distinguish between non-Monet and Monet paintings in the first study, and between Vincent van Gogh and non-van Gogh paintings in the second. The paper also discusses the implications of Artificial Intelligence (AI)-generated pieces of art, briefly delving into the difficulty of identifying if artworks are genuine or not to a computing system. The results show that it is possible to develop a theoretical multi-classifier through the fusion of various binary classifiers, thereby creating an efficient and scalable approach for handling large datasets and many artists. Nonetheless, whereas binary classification proves to be effective for traditional art with respect to accuracy, it cannot differentiate AI paintings imitating artists, thereby representing the limitation of the method. The paper concludes by emphasizing the potential for further advancements in art classification, particularly considering the growing impact of AI-generated artworks.