A Multi-Pass Method for Accelerated Spectral Sampling

Abstract

Spectral Monte Carlo rendering can simulate advanced light phenomena, such as chromatic dispersion, but typically shows a slow convergence behavior. Properly sampling the spectral domain can be challenging in scenes with many complex spectral distributions. To this end, we propose a multi-pass approach. We build and store coarse screen-space estimates of incident spectral radiance and use these to then importance sample the spectral domain. Hereby, we lower variance and reduce noise with little overhead. Our method handles challenging scenarios with difficult spectral distributions, many different emitters, and participating media. Finally, it can be integrated into existing spectral rendering methods for an additional acceleration.

Cite

  @inproceedings{ RuitEisemann2021,
    author    = { Mark van de Ruit and Elmar Eisemann },
    title     = { A Multi-Pass Method for Accelerated Spectral Sampling },
    year      = { 2021 },
    month     = { nov },
    publisher = { Wiley },
    doi       = { 10.1111/cgf.14408 },
    journal   = { Computer Graphics Forum}
  }