Annalena UlschmidORCID iD, Katharina KröslORCID iD, Michael WimmerORCID iD, Bernhard KerblORCID iD
Automated Prioritization for Context-Aware Re-rendering in Editing
SN Computer Science, 6(2025), 2025. [paper] [GitHub]

Information

Abstract

Real-time Monte Carlo path tracing has become a feasible option for interactive 3D scene editing due to recent advancements in GPU ray tracing performance, as well as (AI-accelerated) denoising techniques. While it is thus gaining increasing support in popular modeling software, even minor edits such as adjusting materials or moving small objects typically require current solutions to discard previous samples and restart the image formation process from scratch. A recent solution introduced two adaptive, priority-based re-rendering techniques implementing incremental updates while focusing first on reconstructing regions of high importance and gradually addressing less critical areas. An extensive user study compared these prioritized renderings with conventional same-time re-rendering to evaluate their effectiveness for interactive scene editing. Results indicate a significant preference for incremental rendering techniques for editing small objects over traditional full-screen re-rendering with denoising, even with basic priority policies. Building upon these results, we revisit the underlying design choices and derive more sophisticated priority policies that respect global illumination effects (shadows and reflections) as well as employing attention-based techniques (based either on eye tracking to prioritize areas in the user’s gaze or, alternatively, using the cursor position).

Additional Files and Images

Additional images and videos

image
image

Additional files

paper
paper

Weblinks

BibTeX

@article{ulschmid-2025-apc,
  title =      "Automated Prioritization for Context-Aware Re-rendering in
               Editing",
  author =     "Annalena Ulschmid and Katharina Kr\"{o}sl and Michael Wimmer
               and Bernhard Kerbl",
  year =       "2025",
  abstract =   "Real-time Monte Carlo path tracing has become a feasible
               option for interactive 3D scene editing due to recent
               advancements in GPU ray tracing performance, as well as
               (AI-accelerated) denoising techniques. While it is thus
               gaining increasing support in popular modeling software,
               even minor edits such as adjusting materials or moving small
               objects typically require current solutions to discard
               previous samples and restart the image formation process
               from scratch. A recent solution introduced two adaptive,
               priority-based re-rendering techniques implementing
               incremental updates while focusing first on reconstructing
               regions of high importance and gradually addressing less
               critical areas. An extensive user study compared these
               prioritized renderings with conventional same-time
               re-rendering to evaluate their effectiveness for interactive
               scene editing. Results indicate a significant preference for
               incremental rendering techniques for editing small objects
               over traditional full-screen re-rendering with denoising,
               even with basic priority policies. Building upon these
               results, we revisit the underlying design choices and derive
               more sophisticated priority policies that respect global
               illumination effects (shadows and reflections) as well as
               employing attention-based techniques (based either on eye
               tracking to prioritize areas in the user’s gaze or,
               alternatively, using the cursor position).",
  articleno =  "353",
  doi =        "10.1007/s42979-025-03863-z",
  issn =       "2661-8907",
  journal =    "SN Computer Science",
  number =     "2025",
  pages =      "14",
  volume =     "6",
  publisher =  "Springer Nature",
  keywords =   "Path Tracing, Scene editing, Adaptive rendering, empirical
               studies",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2025/ulschmid-2025-apc/",
}