Visibility precomputation with RTX ray tracing

Information

• Publication Type: Master Thesis
• Workgroup(s)/Project(s):
  • Rendering and Modeling
• Date: March 2021
• Date (Start): 1. March 2020
• Date (End): 9. March 2021
• Diploma Examination: 9. March 2021
• Open Access: yes
• First Supervisor: Michael Wimmer
• Pages: 89
• Keywords: visibitility culling, ray tracing

Abstract

Visibility computation is a common problem in the field of computer graphics. Examples include occlusion culling, where parts of the scene are culled away, or global illumination simulations, which are based on the mutual visibility of pairs of points to calculate lighting. In this thesis, an aggressive from-region visibility technique called Guided Visibility Sampling++ (GVS++) is presented. The proposed technique improves the Guided Visibility Sampling algorithm through improved sampling strategies, thus achieving low error rates on various scenes, and being over four orders of magnitude faster than the original CPU-based Guided Visibility Sampling implementation. We present intelligent sampling strategies that use ray casting to determine a set of triangles visible from a flat or volumetric rectangular region in space. This set is called a potentially visible set (PVS). Based on initial random sampling, subsequent exploration phases progressively grow an intermediate solution. A termination criterion is used to terminate the PVS search. A modern implementation using the Vulkan graphics API and RTX ray tracing is discussed. Furthermore, optimizations are shown that allow for an implementation that is over 20 times faster than a naive implementation.

Additional Files and Images

Weblinks

• publication
  i3D paper from the thesis
• Entry in reposiTUm (TU Wien Publication Database)
• DOI: 10.34726/hss.2021.79729

BibTeX

@mastersthesis{KOCH-2021-GVSDA,
  title =      "Visibility precomputation with RTX ray tracing",
  author =     "Thomas Bernhard Koch",
  year =       "2021",
  abstract =   "Visibility computation is a common problem in the field of
               computer graphics. Examples include occlusion culling, where
               parts of the scene are culled away, or global illumination
               simulations, which are based on the mutual visibility of
               pairs of points to calculate lighting. In this thesis, an
               aggressive from-region visibility technique called Guided
               Visibility Sampling++ (GVS++) is presented. The proposed
               technique improves the Guided Visibility Sampling algorithm
               through improved sampling strategies, thus achieving low
               error rates on various scenes, and being over four orders of
               magnitude faster than the original CPU-based Guided
               Visibility Sampling implementation. We present intelligent
               sampling strategies that use ray casting to determine a set
               of triangles visible from a flat or volumetric rectangular
               region in space. This set is called a potentially visible
               set (PVS). Based on initial random sampling, subsequent
               exploration phases progressively grow an intermediate
               solution. A termination criterion is used to terminate the
               PVS search. A modern implementation using the Vulkan
               graphics API and RTX ray tracing is discussed. Furthermore,
               optimizations are shown that allow for an implementation
               that is over 20 times faster than a naive implementation.",
  month =      mar,
  pages =      "89",
  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",
  keywords =   "visibitility culling, ray tracing",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2021/KOCH-2021-GVSDA/",
}