Image-Space Metaballs Using Deep Learning

Information

• Publication Type: Master Thesis
• Workgroup(s)/Project(s):
  • Visual Computing: Illustrative Visualization
• Date: July 2019
• Date (Start): 2018
• Date (End): 24. July 2019
• First Supervisor: Ivan Viola

Abstract

Metaballs are a type of implicit surface 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 di˙erent approximate methods for rendering metaballs have been developed. In this thesis, the foundations of metaballs and neural networks are discussed, and a new approach to rendering metaballs using Deep Learning that is fast enough for use in real-time applications is presented. The system uses an image-to-image translation approach. For that, first the metaballs are rendered using a very simplified representation to an image. This image is then used as input to a neural network that outputs a depth, normal and base color bu˙er that can be combined using a deferred shading renderer to produce a final image.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@mastersthesis{horvath-2018-ism,
  title =      "Image-Space Metaballs Using Deep Learning",
  author =     "Robert Horvath",
  year =       "2019",
  abstract =   "Metaballs are a type of implicit surface 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 di˙erent
               approximate methods for rendering metaballs have been
               developed. In this thesis, the foundations of metaballs and
               neural networks are discussed, and a new approach to
               rendering metaballs using Deep Learning that is fast enough
               for use in real-time applications is presented. The system
               uses an image-to-image translation approach. For that, first
               the metaballs are rendered using a very simplified
               representation to an image. This image is then used as input
               to a neural network that outputs a depth, normal and base
               color bu˙er that can be combined using a deferred shading
               renderer to produce a final image.",
  month =      jul,
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Research Unit of Computer Graphics, Institute of Visual
               Computing and Human-Centered Technology, Faculty of
               Informatics, TU Wien",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2019/horvath-2018-ism/",
}