Vox2Net: From 3D Shapes to Network Sculptures

Abstract

Sculptures provide the highest degree of physicality in art, allowing the human viewer to make full use of their three dimensional understanding and connect with the piece of art. The advent of Generative Adversarial Networks (GAN) has afforded Artificial Intelligence with a means to create imagery, music and other products which rival human creations. Here we introduce a modified pix2pix GAN, which we call Vox2Vox, that is able to convert different 3D representations of a 3D object to one another. In particular, we teach Vox2Vox to construct a 3D network as an abstract way to represent a sculpture, a construction we call Vox2Net. The input of Vox2Net is a point cloud of the 3D sculpture and its output is spherical nodes and tubular links which together mimic the abstract shape of the original sculpture. Vox2Vox allows the user to convert 3D shapes to any abstract representation of the shape, as well as different styles using the appropriate training data.