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:20260817T171531Z
LOCATION:Hall B5 (1)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241204T133400
DTEND;TZID=Asia/Tokyo:20241204T134600
UID:siggraphasia_SIGGRAPH Asia 2024_sess115_papers_126@linklings.com
SUMMARY:Real-time Large-scale Deformation of Gaussian Splatting
DESCRIPTION:Lin Gao (Institute of Computing Technology, Chinese Academy of
  Sciences; University of Chinese Academy of Sciences); Jie Yang (Institute
  of Computing Technology, Chinese Academy of Sciences); Bo-Tao Zhang, Jia-
 Mu Sun, and Yu-Jie Yuan (Institute of Computing Technology, Chinese Academ
 y of Sciences; University of Chinese Academy of Sciences); Hongbo Fu (Hong
  Kong University of Science and Technology); and Yu-Kun Lai (Cardiff Unive
 rsity)\n\nNeural implicit representations, including Neural Distance Field
 s and Neural Radiance Fields, have demonstrated significant capabilities f
 or reconstructing surfaces with complicated geometry and topology, and gen
 erating novel views of a scene. Nevertheless, it is challenging for users 
 to directly deform or manipulate these implicit representations with large
  deformations in the real-time fashion.Gaussian Splatting (GS) has recentl
 y become a promising method with explicit geometry for representing static
  scenes and facilitating high-quality and real time synthesis of novel vie
 ws. However, it cannot be easily deformed due to the use of discrete Gauss
 ians and lack of explicit topology. To address this, we develop a novel GS
 -based method that enables interactive deformation. Our key idea is to des
 ign an innovative mesh-based GS representation, which is integrated into G
 aussian learning and manipulation. 3D Gaussians are defined over an explic
 it mesh, and they are bound with each other: the rendering of 3D Gaussians
  guides the mesh face split for adaptive refinement, and the mesh face spl
 it directs the splitting of 3D Gaussians. Moreover, the explicit mesh cons
 traints help regularize the Gaussian distribution, suppressing poor-qualit
 y Gaussians (e.g. , misaligned Gaussians, long-narrow shaped Gaussians), t
 hus enhancing visual quality and avoiding artifacts during deformation. Ba
 sed on this representation, we further introduce a large-scale Gaussian de
 formation technique to enable deformable GS, which alters the parameters o
 f 3D Gaussians according to the manipulation of the associated mesh. Our m
 ethod benefits from existing mesh deformation datasets for more realistic 
 data-driven Gaussian deformation. Extensive experiments show that our appr
 oach achieves high-quality reconstruction and effective deformation, while
  maintaining promising rendering results at a high frame rate (65 FPS on a
 verage on a single commodity GPU).\n\nRegistration Category: Full Access, 
 Full Access Supporter\n\nLanguage Format: English Language\n\nSession Chai
 r: Peng-Shuai Wang (Peking University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_126&sess=sess115
END:VEVENT
END:VCALENDAR