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:20250110T023312Z
LOCATION:Hall B5 (1)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241204T132300
DTEND;TZID=Asia/Tokyo:20241204T133400
UID:siggraphasia_SIGGRAPH Asia 2024_sess115_papers_149@linklings.com
SUMMARY:GaussianObject: High-Quality 3D Object Reconstruction from Four Vi
 ews with Gaussian Splatting
DESCRIPTION:Technical Papers\n\nChen Yang and Sikuang Li (Shanghai Jiao To
 ng University), Jiemin Fang (Huawei), Ruofan Liang (University of Toronto)
 , Lingxi Xie and Xiaopeng Zhang (Huawei), Wei Shen (Shanghai Jiao Tong Uni
 versity), and Qi Tian (Huawei)\n\nReconstructing and rendering 3D objects 
 from highly sparse views is of critical importance for promoting applicati
 ons of 3D vision techniques and improving user experience. However, images
  from sparse views only contain very limited 3D information, leading to tw
 o significant challenges: 1) Difficulty in building multi-view consistency
  as images for matching are too few; 2) Partially omitted or highly compre
 ssed object information as view coverage is insufficient. To tackle these 
 challenges, we propose GaussianObject, a framework to represent and render
  the 3D object with Gaussian splatting that achieves high rendering qualit
 y with only 4 input images. We first introduce techniques of visual hull a
 nd floater elimination, which explicitly inject structure priors into the 
 initial optimization process to help build multi-view consistency, yieldin
 g a coarse 3D Gaussian representation. Then we construct a Gaussian repair
  model based on diffusion models to supplement the omitted object informat
 ion, where Gaussians are further refined. We design a self-generating stra
 tegy to obtain image pairs for training the repair model. We further desig
 n a COLMAP-free variant, where pre-given accurate camera poses are not req
 uired, which achieves competitive quality and facilitates wider applicatio
 ns. GaussianObject is evaluated on several challenging datasets, including
  MipNeRF360, OmniObject3D, OpenIllumination, and our-collected unposed ima
 ges, achieving superior performance from only four views and significantly
  outperforming previous SOTA methods.\n\nRegistration Category: Full Acces
 s, Full Access Supporter\n\nLanguage Format: English Language\n\nSession C
 hair: Peng-Shuai Wang (Peking University)
URL:https://asia.siggraph.org/2024/program/?id=papers_149&sess=sess115
END:VEVENT
END:VCALENDAR