Image-Space Metaballs using Deep Learning (DAAV)

Metaballs are implicit surfaces that are used to model organic-looking shapes and fluids. Accurate rendering of three-dimensional Metaballs is typically done using ray-casting, which is computationally expensive and not suitable for real-time applications, therefore different approximate methods for rendering Metaballs have been developed. In this thesis, a new approach to rendering Metaballs efficiently and fast enough for real-time applications using Deep Learning is proposed. A simplified representation of Metaballs is rendered to textures that are then fed to a neural network that outputs multiple depth, normal and base color buffers that are combined using deferred shading to produce an image that resembles the result of a renderer using ray-casting.