Early Renaissance Art Generation Using Deep Convolutional Generative Adversarial Networks
T. Sanjay, J Gladwin, A. G. Sreedevi, M Deepan · 2023
Recent developments in generative art technology have demonstrated the efficacy of neural networks trained on publicly available samples in creating lifelike visuals. However, there are situations where there aren't many training samples to choose from. It's safe to say that you won't see another picture quite like an early Renaissance masterpiece again. In this paper, a pipeline consisting of Deep Convolutional Generative Adversarial Network (DCGAN) preceeded by an augmentation setup is proposed for the automatic generation of early Renaissance-style paintings. Using a 100-dimensional noise vector as input, a neural network with deep convolutional layers is employed as the generator model. The discriminator is a deep convolutional neural network-based classifier with two output classes – authentic and counterfeit. The proposed model uses an open-source dataset of Early Renaissance artwork hand-picked from the WikiArt database. Since original early renaissance paintings are so hard to come by, a transformation process including polarization, inversion, elastic augmentation, solarization, and gaussian noise is employed to generate more samples. The significance of this research is rising high over time due to the scarcity of data and the difficulty in recreating intricate brush strokes in generative models. The investigation is motivated by an interest in Chinese landscape paintings [1] as a source of generative inquiry.