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:20260817T171535Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241205T131100
DTEND;TZID=Asia/Tokyo:20241205T132300
UID:siggraphasia_SIGGRAPH Asia 2024_sess131_papers_1039@linklings.com
SUMMARY:BSDF importance sampling using a diffusion model
DESCRIPTION:Ziyang Fu, Yash Belhe, Haolin Lu, Liwen Wu, Bing Xu, and Tzu-M
 ao Li (University of California San Diego)\n\nMany real-world materials fe
 ature complex BSDFs (Bidirectional Scattering Distribution Functions) that
  require significant storage, making the use of neural networks to represe
 nt BSDFs appealing. Previous neural sampling methods, primarily using anal
 ytical lobe mixtures and normalizing flows, often struggle with specular m
 aterials, particularly at grazing angles. Furthermore, they are limited to
  reflection, and do not handle transmission. Our key observation is that p
 revious normalizing flows impose significant restriction in their network 
 architecture for easy computation of the Jacobian. However, for low-dimens
 ional sampling such as BSDF sampling, the Jacobian computation is not the 
 bottleneck. Therefore, we propose to use diffusion models to importance sa
 mple full BSDFs. Our method has two variants, one for most reflective mate
 rials that learns a distribution on a disk, and the other for extreme spec
 ular reflective materials and full BSDFs, which learns a distribution on a
  sphere. Our equal-time evaluations show that our method outperforms norma
 lizing flows and significantly surpasses them in certain specular material
 s. Additionally, our model provides an expressive and stable neural sampli
 ng method for any complex BSDFs.\n\nRegistration Category: Full Access, Fu
 ll Access Supporter\n\nLanguage Format: English Language\n\nSession Chair:
  Seungyong Lee (POSTECH)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_1039&sess=sess131
END:VEVENT
END:VCALENDAR