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:20241203T150800
DTEND;TZID=Asia/Tokyo:20241203T151900
UID:siggraphasia_SIGGRAPH Asia 2024_sess107_papers_922@linklings.com
SUMMARY:OLAT Gaussians for Generic Relightable Appearance Acquisition
DESCRIPTION:Zhiyi Kuang (State Key Laboratory of CAD&CG, Zhejiang Universi
 ty); Yanchao Yang and Siyan Dong (University of Hong Kong); Jiayue Ma (Sta
 te Key Laboratory of CAD&CG, Zhejiang University); Hongbo Fu (Hong Kong Un
 iversity of Science and Technology); and Youyi Zheng (State Key Laboratory
  of CAD&CG, Zhejiang University)\n\nOne-light-at-a-time (OLAT) images samp
 le a broader range of object appearance changes than images captured under
  constant lighting and are superior as input to object relighting. Althoug
 h existing methods have produced reasonable relighting quality using OLAT 
 images, they utilize surface-like representations, limiting their capacity
  to model volumetric objects, such as furs. Besides, their rendering proce
 ss is time-consuming and still far from being used in real-time. To addres
 s these issues, we propose OLAT Gaussians to build relightable representat
 ions of objects from multi-view OLAT images. We build our pipeline on 3D G
 aussian Splatting (3DGS), which achieves real-time high-quality rendering.
  To augment 3DGS with relighting capability, we assign each Gaussian a lea
 rnable feature vector, serving as an index to query the objects’ appearanc
 e field. Specifically, we decompose the appearance field into light transp
 ort and scattering functions. The former accounts for light transmittance 
 and foreshortening effects, while the latter represents the object’s mater
 ial properties to scatter light. Rather than using an off-the-shelf physic
 ally-based parametric rendering formulation, we model both functions using
  multi-layer perceptrons (MLPs). This makes our method suitable for variou
 s objects, e.g., opaque surfaces, semi-transparent volumes, furs, fabrics,
  etc. Given a camera view and a point light position, we compute each Gaus
 sian’s color as the product of the light transport value, the scattering v
 alue, and the light intensity, and then render the target image through th
 e 3DGS rasterizer. To enhance rendering quality, we further utilize a prox
 y mesh to provide OLAT Gaussians with normals to improve highlights and vi
 sibility cues to improve shadows. Extensive experiments demonstrate that o
 ur method produces state-of-the-art rendering quality with significantly m
 ore details in texture-rich areas than previous methods. Our method also a
 chieves real-time rendering, allowing users to interactively modify views 
 and lights to get immediate rendering results, which are not available fro
 m the offline rendering of previous methods.\n\nRegistration Category: Ful
 l Access, Full Access Supporter\n\nLanguage Format: English Language\n\nSe
 ssion Chair: Hongzhi Wu (Zhejiang University; State Key Laboratory of CAD&
 CG, Zhejiang University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_922&sess=sess107
END:VEVENT
END:VCALENDAR