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:20260817T171533Z
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:Jiadi Cui (ShanghaiTech University, Stereye Inc.); Junming Cao
  (Shanghai Advanced Research Institute, Chinese Academy of Sciences; Unive
 rsity of Chinese Academy of Sciences); Fuqiang Zhao (ShanghaiTech Universi
 ty, NeuDim Inc.); Zhipeng He and Yifan Chen (ShanghaiTech University); Yuh
 ui Zhong (DGene Inc.); Lan Xu and Yujiao Shi (ShanghaiTech University); Yi
 ngliang Zhang (DGene Inc.); and Jingyi Yu (ShanghaiTech University)\n\nLar
 ge garages are ubiquitous yet intricate scenes that present unique challen
 ges due to their monotonous colors, repetitive patterns, reflective surfac
 es, and transparent vehicle glass. Conventional Structure from Motion (SfM
 ) methods for camera pose estimation and 3D reconstruction often fail in t
 hese environments due to poor correspondence construction. To address thes
 e challenges, we introduce LetsGo, a LiDAR-assisted Gaussian splatting fra
 mework for large-scale garage modeling and rendering.\nWe develop a handhe
 ld scanner, Polar, equipped with IMU, LiDAR, and a fisheye camera, to faci
 litate accurate data acquisition. Using this Polar device, we present the 
 GarageWorld dataset, consisting of eight expansive garage scenes with dive
 rse geometric structures, which will be made publicly available for furthe
 r research.\nOur approach demonstrates that LiDAR point clouds collected b
 y the Polar device significantly enhance a suite of 3D Gaussian splatting 
 algorithms for garage scene modeling and rendering. We introduce a novel d
 epth regularizer that effectively eliminates floating artifacts in rendere
 d images.\nAdditionally, we propose a multi-resolution 3D Gaussian represe
 ntation designed for Level-of-Detail (LOD) rendering. This includes adapte
 d scaling factors for individual levels and a random-resolution-level trai
 ning scheme to optimize the Gaussians across different resolutions. This r
 epresentation enables efficient rendering of large-scale garage scenes on 
 lightweight devices via a web-based renderer.\nExperimental results on our
  GarageWorld dataset, as well as on ScanNet++ and KITTI-360, demonstrate t
 he superiority of our method in terms of rendering quality and resource ef
 ficiency.\n\nRegistration Category: Full Access, Full Access Supporter\n\n
 Language Format: English Language\n\nSession Chair: Bernhard Kerbl (Techni
 cal University of Vienna)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_170&sess=sess104
END:VEVENT
END:VCALENDAR