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:20260817T171532Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241204T163000
DTEND;TZID=Asia/Tokyo:20241204T164100
UID:siggraphasia_SIGGRAPH Asia 2024_sess122_papers_630@linklings.com
SUMMARY:InstanceTex: Instance-level Controllable Texture Synthesis for 3D 
 Scenes via Diffusion Priors
DESCRIPTION:Mingxin Yang (Shenzhen Institute of Advanced Technology, Chine
 se Academy of Sciences); Jianwei Guo (Institute of Automation, Chinese Aca
 demy Of Sciences); Yuzhi Chen (School of Artificial Intelligence, Universi
 ty of Chinese Academy of Sciences); Lan Chen (Institute of Automation, Chi
 nese Academy of Sciences); Pu Li (Institute of Automation, Chinese Academy
  Of Sciences); Zhanglin Cheng (Shenzhen Institute of Advanced Technology, 
 Chinese Academy of Sciences); Xiaopeng Zhang (Institute of Automation, Chi
 nese Academy Of Sciences); and Hui Huang (Shenzhen University (SZU))\n\nAu
 tomatically generating high-fidelity texture for a complex scene remains a
 n open problem in computer graphics. While pioneering text-to-texture work
 s based on 2D diffusion models have achieved fascinating results on single
  objects, they either suffer from style inconsistency and semantic misalig
 nment or require extensive optimization time/memory when scaling it up to 
 a large scene. To address these challenges, we introduce InstanceTex, a no
 vel method to synthesize realistic and style-consistent textures for large
 -scale scenes. At its core, InstanceTex proposes an instance-level control
 lable texture synthesis approach based on an instance layout representatio
 n, which enables precise control over the instances while keeping the glob
 al style consistency. We also propose a local synchronized multi-view diff
 usion approach to enhance local texture consistency by sharing the latent 
 denoised content among neighboring views in a mini-batch. Finally, tailore
 d to scene texture mapping, we develop Neural MipTexture inspired by the M
 ipmaps to reduce the aliasing artifacts. Extensive texturing results on in
 door and outdoor scenes show that InstanceTex produces high-quality and co
 nsistent textures with the superior quality compared to prior texture gene
 ration alternatives.\n\nRegistration Category: Full Access, Full Access Su
 pporter\n\nLanguage Format: English Language\n\nSession Chair: Minhyuk Sun
 g (Korea Advanced Institute of Science and Technology (KAIST))\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_630&sess=sess122
END:VEVENT
END:VCALENDAR