SEED Research & Announcements Blogs Publications Open Source Careers Contact Us Research & Announcements Blogs Publications Open Source Careers Contact Us

Beyond White Noise for Real-Time Rendering

Going beyond white noise for temporal and spatial denoising in real-time rendering can produce better results with no increase in rendering time.

In this presentation, SEED’s Alan Wolfe discusses the use of different types of noise for random number generation, focusing on applications in real time rendering, and includes research just published at I3D 2024.

In randomized rendering algorithms, white noise is used by the supporting math and does indeed work. However, changing the source of random numbers can produce much higher-quality results without increasing rendering time. So, without even switching the rendering algorithm, using better noise can significantly improve the perceptual quality or measured error of images in ray trace rendering and beyond.

Alan’s presentation covers:

  • Randomness and fairness in number generation
  • Stochastic rendering
  • Noise textures and error patterns

Download the presentation deck (PDF 14 MB).

 

Related News

Incorporating ML Research Into Audio Production: ExFlowSions Case Study

SEED
Jun 25, 2024
Mónica Villanueva and Jorge García present the challenges and lessons learned from turning a machine learning generative model from a research project into a game production tool.

Evaluating Gesture Generation in a Large-Scale Open Challenge

SEED
May 9, 2024
This paper, published in Transactions on Graphics, reports on the second GENEA Challenge, a project to benchmark data-driven automatic co-speech gesture generation.

Filter-Adapted Spatio-Temporal Sampling for Real-Time Rendering

SEED
May 1, 2024
This paper, presented at I3D 2024, discusses how to tailor the frequencies of rendering noise to improve image denoising in real time rendering.