3D Image Generation Using Generative Adversarial Network for Virtual Art Gallery

Abstract

Gallery art websites are often used to display artistic work for user accessibility. The conventional web-based gallery is built on the hyper-text Markup Language (HTML) which is limited in terms of content dynamism, interactivity, and scalability. GANs provide advanced capabilities with continuously evolving art experiences, making them a powerful tool for modern digital art galleries. Artistic websites have significantly contributed to the exhibition of artistic images since they can be accessed anywhere. However, the artistic web suffers from being too passive and lacks in-depth interactivity to keep people meaningfully engaged with an exhibition virtually. This paper explores the exhibition of 3D images within a virtual art gallery using an intelligent artistic web-based applications framework that integrates Variational Autoencoder and 3-D Signed Distance Function Cycle GAN (VAE-3DSDFCycleGAN) and quantitative questionnaire methods. The virtual gallery utilises GAN architectures to produce diverse and original 3D artworks, addressing traditional art galleries' spatial, viewing dimensions, image quality, and accessibility limitations. The questionnaire was used to evaluate the user’s satisfaction. The experiment was done on the Coco African Mask dataset to generate 3-D images, yielding a high result and satisfaction in terms of the ease of use and viewing of the artistic image contents.