BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Hong_Kong
X-LIC-LOCATION:Asia/Hong_Kong
BEGIN:STANDARD
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
TZNAME:HKT
DTSTART:19911015T033000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20251218T030657Z
LOCATION:Meeting Room S426+S427\, Level 4
DTSTART;TZID=Asia/Hong_Kong:20251218T111200
DTEND;TZID=Asia/Hong_Kong:20251218T112300
UID:siggraphasia_SIGGRAPH Asia 2025_sess152_papers_2060@linklings.com
SUMMARY:JumpingGS: Level-jump 3D Gaussian Representation for Delicate Text
 ures in Aerial Large-scale Scene Rendering
DESCRIPTION:Jiongming Qin (Wuhan University, School of Computer Science); 
 Kaixuan Zhou (Huawei Technologies, Riemann Lab); and Yu Jiang, Huizhi Zhu,
  Fei Luo, and Chunxia Xiao (Wuhan University, School of Computer Science)\
 n\nExisting 3D Gaussian (3DGS) based methods tend to produce blurriness an
 d artifacts on delicate textures (small objects and high-frequency texture
 s) in aerial large-scale scenes. The reason is that the delicate textures 
 usually occupy a relatively small number of pixels, and the accumulated gr
 adients from loss function are difficult to promote the splitting of 3DGS.
  To minimize the rendering error, the model will use a small number of lar
 ge Gaussians to cover these details, resulting in blurriness and artifacts
 . To solve the above problem, we propose a novel hierarchical Gaussian: Ju
 mpingGS. JumpingGS assigns different levels to Gaussians to establish a hi
 erarchical representation. Low-level Gaussians are responsible for the coa
 rse appearance, while high-level Gaussians are responsible for the details
 . First, we design a splitting strategy that allows low-level Gaussians to
  skip intermediate levels and directly split the appropriate high-level Ga
 ussians for delicate textures. This level-jump splitting ensures that the 
 weak gradients of delicate textures can always activate a higher level ins
 tead of being ignored by the intermediate levels. Second, JumpingGS reduce
 s the gradient and opacity thresholds for density control according to the
  representation levels, which improves the sensitivity of high-level Gauss
 ians to delicate textures. Third, we design a novel training strategy to d
 etect training views in hard-to-observe regions, and train the model multi
 ple times on these views to alleviate underfitting. Experiments on aerial 
 large-scale scenes demonstrate that JumpingGS outperforms existing 3DGS-ba
 sed methods, accurately and efficiently recovering delicate textures in la
 rge scenes.\n\nRegistration Category: Full Access, Full Access Supporter\n
 \nSession Chair: Lin Gao (University of Chinese Academy of Sciences)\n\n
END:VEVENT
END:VCALENDAR