Adversarial Monte Carlo Denoising with Conditioned Auxiliary Feature Modulation

BING XU1             JUNFEI ZHANG1             RUI WANG2             KUN XU3             YONG-LIANG YANG4             CHUAN LI5             RUI TANG1

1KooLab, Kujiale, China                                          3BNRist, Department of Computer Science and Technology, Tsinghua University, China
2State Key Laboratory of CAD & CG, Zhejiang University, China                      4University of Bath, UK                      5Lambda Labs Inc, USA


Abstract

Denoising Monte Carlo rendering with a very low sample rate remains a major challenge in the photo-realistic rendering research. Many previous works, including regression-based and learning-based methods, have been explored to achieve better rendering quality with less computational cost. However, most of these methods rely on handcrafted optimization objectives, which lead to artifacts such as blurs and unfaithful details. In this paper, we present an adversarial approach for denoising Monte Carlo rendering. Our key insight is that generative adversarial networks can help denoiser networks to produce more realistic high-frequency details and global illumination by learning the distribution from a set of high-quality Monte Carlo path tracing images.We also adapt a novel feature modulation method to utilize auxiliary features better, including normal, albedo and depth. Compared to previous state-of-the-art methods, our approach produces a better reconstruction of the Monte Carlo integral from a few samples, performs more robustly at different sample rates, and takes only a second for megapixel images.


Paper

Paper | Paper (low res) | Supplemental Material

@article {xuMCGANsa2019,
author = {Bing Xu and Junfei Zhang and Rui Wang and Kun Xu and Yong-Liang Yang and Chuan Li and Rui Tang},
title = {Adversarial Monte Carlo Denoising with Conditioned Auxiliary Feature},
journal = {ACM Transactions on Graphics (Proceedings of ACM SIGGRAPH Asia 2019)},
year = {2019},
volume = {38},
pages = {224:1--224:12},
number = {6},
}


Code & Dataset

Our code and model weights are released here. The large scale indoor dataset from Kujiale.com is published on both Google cloud drive and Badiu cloud drive with code wahy. Please choose cloud drive whichever is more convenient.


Interactive Viewer

Use the interactive viewer to inspect the image details of our methods.