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:20260817T171530Z
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:Chen Yang and Sikuang Li (Shanghai Jiao Tong University), Jiem
 in Fang (Huawei), Ruofan Liang (University of Toronto), Lingxi Xie and Xia
 openg Zhang (Huawei), Wei Shen (Shanghai Jiao Tong University), and Qi Tia
 n (Huawei)\n\nReconstructing and rendering 3D objects from highly sparse v
 iews is of critical importance for promoting applications of 3D vision tec
 hniques and improving user experience. However, images from sparse views o
 nly contain very limited 3D information, leading to two significant challe
 nges: 1) Difficulty in building multi-view consistency as images for match
 ing are too few; 2) Partially omitted or highly compressed object informat
 ion as view coverage is insufficient. To tackle these challenges, we propo
 se GaussianObject, a framework to represent and render the 3D object with 
 Gaussian splatting that achieves high rendering quality with only 4 input 
 images. We first introduce techniques of visual hull and floater eliminati
 on, which explicitly inject structure priors into the initial optimization
  process to help build multi-view consistency, yielding a coarse 3D Gaussi
 an representation. Then we construct a Gaussian repair model based on diff
 usion models to supplement the omitted object information, where Gaussians
  are further refined. We design a self-generating strategy to obtain image
  pairs for training the repair model. We further design a COLMAP-free vari
 ant, where pre-given accurate camera poses are not required, which achieve
 s competitive quality and facilitates wider applications. GaussianObject i
 s evaluated on several challenging datasets, including MipNeRF360, OmniObj
 ect3D, OpenIllumination, and our-collected unposed images, achieving super
 ior performance from only four views and significantly outperforming previ
 ous SOTA methods.\n\nRegistration Category: Full Access, Full Access Suppo
 rter\n\nLanguage Format: English Language\n\nSession Chair: Peng-Shuai Wan
 g (Peking University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_149&sess=sess115
END:VEVENT
END:VCALENDAR