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:20250110T023309Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241203T134200
DTEND;TZID=Asia/Tokyo:20241203T135600
UID:siggraphasia_SIGGRAPH Asia 2024_sess104_papers_170@linklings.com
SUMMARY:LetsGo: Large-Scale Garage Modeling and Rendering via LiDAR-Assist
 ed Gaussian Primitives
DESCRIPTION:Technical Papers\n\nJiadi Cui (ShanghaiTech University, Sterey
 e Inc.); Junming Cao (Shanghai Advanced Research Institute, Chinese Academ
 y of Sciences; University of Chinese Academy of Sciences); Fuqiang Zhao (S
 hanghaiTech University, NeuDim Inc.); Zhipeng He and Yifan Chen (ShanghaiT
 ech University); Yuhui Zhong (DGene Inc.); Lan Xu and Yujiao Shi (Shanghai
 Tech University); Yingliang Zhang (DGene Inc.); and Jingyi Yu (ShanghaiTec
 h University)\n\nLarge garages are ubiquitous yet intricate scenes that pr
 esent unique challenges due to their monotonous colors, repetitive pattern
 s, reflective surfaces, and transparent vehicle glass. Conventional Struct
 ure from Motion (SfM) methods for camera pose estimation and 3D reconstruc
 tion often fail in these environments due to poor correspondence construct
 ion. To address these challenges, we introduce LetsGo, a LiDAR-assisted Ga
 ussian splatting framework for large-scale garage modeling and rendering.\
 nWe develop a handheld scanner, Polar, equipped with IMU, LiDAR, and a fis
 heye camera, to facilitate accurate data acquisition. Using this Polar dev
 ice, we present the GarageWorld dataset, consisting of eight expansive gar
 age scenes with diverse geometric structures, which will be made publicly 
 available for further research.\nOur approach demonstrates that LiDAR poin
 t clouds collected by the Polar device significantly enhance a suite of 3D
  Gaussian splatting algorithms for garage scene modeling and rendering. We
  introduce a novel depth regularizer that effectively eliminates floating 
 artifacts in rendered images.\nAdditionally, we propose a multi-resolution
  3D Gaussian representation designed for Level-of-Detail (LOD) rendering. 
 This includes adapted scaling factors for individual levels and a random-r
 esolution-level training scheme to optimize the Gaussians across different
  resolutions. This representation enables efficient rendering of large-sca
 le garage scenes on lightweight devices via a web-based renderer.\nExperim
 ental results on our GarageWorld dataset, as well as on ScanNet++ and KITT
 I-360, demonstrate the superiority of our method in terms of rendering qua
 lity and resource efficiency.\n\nRegistration Category: Full Access, Full 
 Access Supporter\n\nLanguage Format: English Language\n\nSession Chair: Be
 rnhard Kerbl (Technical University of Vienna)
URL:https://asia.siggraph.org/2024/program/?id=papers_170&sess=sess104
END:VEVENT
END:VCALENDAR