Automatic Feature Selection for Denoising Volumetric Renderings
We propose a method for constructing feature sets that significantly improve the quality of neural denoisers for Monte Carlo renderings with volumetric content.
April 7, 2022
Eurographics Symposium on Rendering (EGSR) (2022)
Xianyao Zhang (DisneyResearch|Studios/ETH Joint Ph.D.)
Melvin Ott (DisneyResearch|Studios/ETH Joint M.Sc.)
Marco Manzi (DisneyResearch|Studios)
Markus Gross (DisneyResearch|Studios/ETH Zurich)
Marios Papas (DisneyResearch|Studios)
We propose a method for constructing feature sets that significantly improve the quality of neural denoisers for Monte Carlo renderings with volumetric content. Starting from a large set of hand-crafted features, we propose a feature selection process to identify significantly pruned near-optimal subsets. While a naive approach would require training and testing a separate denoiser for every possible feature combination, our selection process requires training of only a single probe denoiser for the selection task. Moreover, our approximate solution has an asymptotic complexity that is quadratic to the number of features compared to the exponential complexity of the naive approach, while also producing near-optimal solutions. We demonstrate the usefulness of our approach to various state-of-the-art denoising methods for volumetric content. We observe improvements in denoising quality when using our automatically selected feature sets over the hand-crafted sets proposed by the original methods.