Boosting Histogram-Based Denoising Methods with GPU Optimizations

Sebastian Szeracki, Thorsten Roth, André Hinkenjann, Yongmin Li
12. Workshop Virtuelle Realität und Augmented Reality der GI-Fachgruppe VR/AR - September 2015
Download the publication : SRHL15.pdf [3.1Mo]  


We propose a high-performance GPU implementation of Ray Histogram Fusion (RHF), a denoising method for stochastic global illumination rendering. Based on the CPU implementation of the original algorithm, we present a naive GPU implementation and the necessary optimization steps. Eventually, we show that our optimizations increase the performance of RHF by two orders of magnitude when compared to the original CPU implementation and one order of magnitude compared to the naive GPU implementation. We show how the quality for identical rendering times relates to unfiltered path tracing and how much time is needed to achieve identical quality when compared to an unfiltered path traced result. Finally, we summarize our work and describe possible future applications and research based on this.

Images and movies


BibTex references

  author       = {Szeracki, Sebastian and Roth, Thorsten and Hinkenjann, Andr{\'e} and Li, Yongmin},
  title        = {Boosting Histogram-Based Denoising Methods with GPU Optimizations},
  booktitle    = {12. Workshop Virtuelle Realit{\"a}t und Augmented Reality der GI-Fachgruppe VR/AR},
  month        = {September},
  year         = {2015},
  publisher    = {Shaker Verlag},

Other publications in the database

» Thorsten Roth
» André Hinkenjann
» Yongmin Li