Neural Kernel Regression for Consistent Monte Carlo Denoising

doi:10.1145/3687949

	Neural Kernel Regression for Consistent Monte Carlo Denoising
	Wang, Qi 1; Qiao, Pengju 2; Huo, Yuchi 1,3; Zhai, Shiji 4; Xie, Zixuan 3; Hua, Wei 3; Bao, Hujun 1; Liu, Tao 5
	2024-12-01
发表期刊	ACM TRANSACTIONS ON GRAPHICS
ISSN	0730-0301
卷号	43 期号:6 页码:14
摘要	Unbiased Monte Carlo path tracing that is extensively used in realistic rendering produces undesirable noise, especially with low samples per pixel (spp). Recently, several methods have coped with this problem by importing unbiased noisy images and auxiliary features to neural networks to either predict a fixed-sized kernel for convolution or directly predict the denoised result. Since it is impossible to produce arbitrarily high spp images as the training dataset, the network-based denoising fails to produce high-quality images under high spp. More specifically, network-based denoising is inconsistent and does not converge to the ground truth as the sampling rate increases. On the other hand, the post-correction estimators yield a blending coefficient for a pair of biased and unbiased images influenced by image errors or variances to ensure the consistency of the denoised image. As the sampling rate increases, the blending coefficient of the unbiased image converges to 1, that is, using the unbiased image as the denoised results. However, these estimators usually produce artifacts due to the difficulty of accurately predicting image errors or variances with low spp. To address the above problems, we take advantage of both kernel-predicting methods and post-correction denoisers. A novel kernel-based denoiser is proposed based on distribution-free kernel regression consistency theory, which does not explicitly combine the biased and unbiased results but constrain the kernel bandwidth to produce consistent results under high spp. Meanwhile, our kernel regression method explores bandwidth optimization in the robust auxiliary feature space instead of the noisy image space. This leads to consistent high-quality denoising at both low and high spp. Experiment results demonstrate that our method outperforms existing denoisers in accuracy and consistency.
关键词	Kernel regression Monte Carlo render- ing learning-based denoising
DOI	10.1145/3687949
收录类别	SCI
语种	英语
资助项目	NSFC[62441205] ; National Key R&D Program of China[2023YFF0905102] ; National Key R&D Program of China[2024YDLN0011] ; Key R&D Program of Zhejiang Province[2023C01039]
WOS研究方向	Computer Science
WOS类目	Computer Science, Software Engineering
WOS记录号	WOS:001367474400001
出版者	ASSOC COMPUTING MACHINERY
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/41153
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Huo, Yuchi
作者单位	1.Zhejiang Univ, State Key Lab CAD&CG, Hangzhou, Peoples R China 2.Chinese Acad Sci, Inst Software, Beijing, Peoples R China 3.Zhejiang Lab, Hangzhou, Peoples R China 4.Chinese Acad Sci, Inst Comp Technol, Beijing, Peoples R China 5.Shanghai Maritime Univ, Coll Transport & Commun, Shanghai, Peoples R China
推荐引用方式 GB/T 7714	Wang, Qi,Qiao, Pengju,Huo, Yuchi,et al. Neural Kernel Regression for Consistent Monte Carlo Denoising[J]. ACM TRANSACTIONS ON GRAPHICS,2024,43(6):14.
APA	Wang, Qi.,Qiao, Pengju.,Huo, Yuchi.,Zhai, Shiji.,Xie, Zixuan.,...&Liu, Tao.(2024).Neural Kernel Regression for Consistent Monte Carlo Denoising.ACM TRANSACTIONS ON GRAPHICS,43(6),14.
MLA	Wang, Qi,et al."Neural Kernel Regression for Consistent Monte Carlo Denoising".ACM TRANSACTIONS ON GRAPHICS 43.6(2024):14.