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:20260817T171530Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241205T094200
DTEND;TZID=Asia/Tokyo:20241205T095600
UID:siggraphasia_SIGGRAPH Asia 2024_sess125_papers_443@linklings.com
SUMMARY:Spatiotemporal Bilateral Gradient Filtering for Inverse Rendering
DESCRIPTION:Wesley Chang, Xuanda Yang, Yash Belhe, Ravi Ramamoorthi, and T
 zu-Mao Li (University of California San Diego)\n\nIn inverse rendering, gr
 adient-based methods, which have seen great progress in the recent years, 
 are typically used in conjunction with the Adam optimizer. While Adam usua
 lly improves convergence by temporally filtering gradients over previous i
 terations to reduce noise, it is not tailored to inverse rendering where t
 he target signals (textures, volumes, or geometry) are usually piecewise s
 mooth. Previous work has applied the inverse Laplacian operator to smooth 
 gradients spatially, but this isotropic filtering can often lead to oversm
 oothing. We propose a spatiotemporal optimizer that can significantly spee
 dup the convergence over Adam, by enforcing the optimization parameter upd
 ates to be piecewise smooth through a lightweight spatial domain cross-bil
 ateral filter. We discuss different options of combining spatial filtering
  and Adam's temporal filtering, and provide intuitions for different scena
 rios. We show that our filtering leads to significantly higher-quality rec
 onstructions in different inverse problems including texture, volume and g
 eometry recovery.\n\nRegistration Category: Full Access, Full Access Suppo
 rter\n\nLanguage Format: English Language\n\nSession Chair: Wenzel Jakob (
 École Polytechnique Féderale de Lausanne (EPFL), NVIDIA)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_443&sess=sess125
END:VEVENT
END:VCALENDAR