Real-time Rendering of Indirectly Visible Caustics

Pierre Moreau
Lund University

Michael Doggett
Lund University

GRAPP, February 2022

Abstract

Caustics are a challenging light transport phenomenon to render in real-time, and most previous approaches have used screen-space accumulation based on the fast rasterization hardware of GPUs. This limits the position of photon collection points to first hit screen space locations, and leads to missing caustics that should be visible in a mirror's reflection. In this paper we propose an algorithm that can render caustics visible via specular bounces in real-time. The algorithm takes advantage of hardware-accelerated ray tracing support found in modern GPUs. By constructing an acceleration structure around multiple bounce view ray hit points in world space, and tracing multiple bounce light rays through the scene, we ensure caustics can be created anywhere in the scene, not just in screen space. We analyze the performance and image quality of our approach, and show that we can produce indirectly visible caustics at real-time rates.

News

  • 2022-02-16: Update BibTeX entry with DOI, add author copy of the paper and the presentation slides.
  • 2022-01-20: Add BibTeX entry.
  • 2021-12-03: Add link to GitHub repository.
  • 2021-11-29: Initial upload.

Downloads

Paper
BibTeX entry
@inproceedings{ Moreau2022,
  author        = {Moreau, Pierre and Doggett, Michael},
  title         = {Real-time Rendering of Indirectly Visible Caustics},
  pages         = {39–48},
  DOI           = {10.5220/0010777800003124},

  editor        = {Sousa, A. Augusto and Debattista, Kurt and Bouatouch, Kadi},
  publisher     = {{SciTePress}},
  address       = {Setúbal, {PRT}},
  organization  = {{INSTICC}},
  booktitle     = {Proceedings of the 17th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications},
  series        = {{GRAPP}},
  volume        = {1},
  year          = {2022},
  ISSN          = {2184-4321},
  ISBN          = {978-989-758-555-5},

  eventdate     = {2022-02-06–2022-02-08}
}
Slides
Presentation at GRAPP 2022
Code
https://github.com/LUGGPublic/Indirectly-Visible-Caustics
Videos

Videos

All videos were captured in real-time on a GeForce RTX 3080 GPU (with 10 GB of VRAM).

Animated camera, rendered with 10242 light paths of up to 6 segments.
Animated glass sphere, rendered with 10242 light paths of up to 6 segments.
Animated mesh light, rendered with 10242 light paths of up to 6 segments.
Animated camera, glass sphere, and mesh light, rendered with 10242 light paths of up to 6 segments.