BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250110T023312Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241205T095600
DTEND;TZID=Asia/Tokyo:20241205T101000
UID:siggraphasia_SIGGRAPH Asia 2024_sess125_papers_634@linklings.com
SUMMARY:Neural Kernel Regression for Consistent Monte Carlo Denoising
DESCRIPTION:Technical Papers\n\nQi Wang (State Key Laboratory of CAD&CG, Z
 hejiang University); Pengju Qiao (Institute of Software Chinese Academy of
  Sciences; State Key Laboratory of CAD&CG, Zhejiang University); Yuchi Huo
  (State Key Laboratory of CAD&CG, Zhejiang University; Zhejiang University
 ); Shiji Zhai (Institute of Computing Technology, Chinese Academy of Scien
 ces); Zixuan Xie (Institute of Computing Technology, Chinese Academy of Sc
 iences; Zhejiang Lab); Rengan Xie (State Key Laboratory of CAD&CG, Zhejian
 g University); Wei Hua (Zhejiang Lab); Hujun Bao (State Key Laboratory of 
 CAD&CG, Zhejiang University; Zhejiang University); and Tao Liu (Shanghai M
 aritime University, College of Transport & Communications)\n\nUnbiased Mon
 te Carlo path tracing that is extensively used in realistic rendering prod
 uces undesirable noise, especially with low samples per pixel (spp). Recen
 tly, several methods have coped with this problem by importing unbiased no
 isy images and auxiliary features to neural networks to either predict a f
 ixed-sized kernel for convolution or directly predict the denoised result.
  However, since it is impossible to produce arbitrarily high spp images as
  the training dataset, the network-based denoising fails to produce high-q
 uality images under high spp. More specifically, network-based denoising i
 s not consistent 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 ima
 ge. As the sampling rate increases, the blending coefficient of the unbias
 ed image converges to 1, that is, using the unbiased image as the denoised
  results. However, due to the difficulty of accurately predicting image er
 rors or variances with low spp, these estimators usually produce artifacts
 . To address the above problems, we take advantage of both kernel-predicti
 ng methods and post-correction denoisers. Specifically, we propose a novel
  kernel-based denoiser based on distribution-free kernel regression consis
 tency theory, which does not explicitly combine the biased and unbiased re
 sults but constrains the kernel bandwidth to produce consistent results un
 der high spp. Meanwhile, our kernel regression method explores the bandwid
 th optimization in the robust auxiliary feature space instead of the noisy
  image space, which leads to consistent high-quality denoising at both low
  and high spp. Experiments demonstrate that our method outperforms existin
 g denoisers in accuracy and consistency.\n\nRegistration Category: Full Ac
 cess, Full Access Supporter\n\nLanguage Format: English Language\n\nSessio
 n Chair: Wenzel Jakob (École Polytechnique Fédérale de Lausanne)
URL:https://asia.siggraph.org/2024/program/?id=papers_634&sess=sess125
END:VEVENT
END:VCALENDAR