Implementing Probabilistic Connections for Bidirectional Path Tracing in the Mitsuba Renderer

Information

• Publication Type: Bachelor Thesis
• Workgroup(s)/Project(s):
  • Rendering and Modeling
• Date: September 2017
• Date (Start): 23. May 2017
• Date (End): 3. September 2017
• Matrikelnummer: 01328690
• First Supervisor:
  • Hiroyuki Sakai
  • Michael Wimmer
• Keywords: physically based rendering, Monte Carlo rendering, bidirectional path tracing, probabilistic connections for bidirectional path tracing, Mitsuba, importance sampling

Abstract

Light transport simulation algorithms are remarkably adept at recreating a large variety of light phenomena which occur in nature. As such they have seen widespread adoption across the industry, which made it paramount to create efficient and robust algorithms. One recent algorithm which tries to deal with this problem is known as Probabilistic Connections for Bidirectional Path Tracing (PCBPT). It builds upon the classical Bidirectional Path Tracing (BDPT) algorithm. In Bidirectional Path Tracing, a ray is traced from the sensor as well as from the emitter. The two rays are then connected to calculate the light contribution to image pixels. PCBPT extends this idea to support connecting multiple emitter paths to one sensor subpath, and introduces importance sampling as a way of choosing the most suitable emitter paths. Unfortunately, there was no implementation of PCBPT publically available, which is why we implemented it into the open-source Mitsuba renderer. We evaluate the algorithm against standard BDPT on a variety of different scenes. Our comparisons provide insight into what type of scenes PCBPT can help improve and where the additional computational cost presents too much of an overhead.

Additional Files and Images

Weblinks

• Author's Web Page

BibTeX

@bachelorsthesis{dodik-2017-pcbpt,
  title =      "Implementing Probabilistic Connections for Bidirectional
               Path Tracing in the Mitsuba Renderer",
  author =     "Nikola Dodik",
  year =       "2017",
  abstract =   "Light transport simulation algorithms are remarkably adept
               at recreating a large variety of light phenomena which occur
               in nature. As such they have seen widespread adoption across
               the industry, which made it paramount to create efficient
               and robust algorithms. One recent algorithm which tries to
               deal with this problem is known as Probabilistic Connections
               for Bidirectional Path Tracing (PCBPT). It builds upon the
               classical Bidirectional Path Tracing (BDPT) algorithm. In
               Bidirectional Path Tracing, a ray is traced from the sensor
               as well as from the emitter. The two rays are then connected
               to calculate the light contribution to image pixels. PCBPT
               extends this idea to support connecting multiple emitter
               paths to one sensor subpath, and introduces importance
               sampling as a way of choosing the most suitable emitter
               paths. Unfortunately, there was no implementation of PCBPT
               publically available, which is why we implemented it into
               the open-source Mitsuba renderer. We evaluate the algorithm
               against standard BDPT on a variety of different scenes. Our
               comparisons provide insight into what type of scenes PCBPT
               can help improve and where the additional computational cost
               presents too much of an overhead.",
  month =      sep,
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology ",
  keywords =   "physically based rendering, Monte Carlo rendering,
               bidirectional path tracing, probabilistic connections for
               bidirectional path tracing, Mitsuba, importance sampling",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2017/dodik-2017-pcbpt/",
}