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:20250110T023313Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241206T144500
DTEND;TZID=Asia/Tokyo:20241206T145600
UID:siggraphasia_SIGGRAPH Asia 2024_sess149_papers_967@linklings.com
SUMMARY:Boosting 3D Object Generation through PBR Materials
DESCRIPTION:Technical Papers\n\nYitong Wang (Fudan University, Shanghai Ar
 tificial Intelligence Laboratory); Xudong Xu (Shanghai Artificial Intellig
 ence Laboratory); Li Ma (Scanline VFX Studio); Haoran Wang (Shanghai Jiaot
 ong University); and Bo Dai (Shanghai Artificial Intelligence Laboratory)\
 n\nAutomatic 3D content creation has gained increasing attention recently,
  due to its potential in various applications such as video games, film in
 dustry, and AR/VR. \nRecent advancements in diffusion models and multimoda
 l models have notably improved the quality and efficiency of 3D object gen
 eration given a single RGB image. \nHowever, 3D objects generated even by 
 state-of-the-art methods are still unsatisfactory compared to human-create
 d assets. \nConsidering only textures instead of materials makes these met
 hods encounter challenges in photo-realistic rendering, relighting, and fl
 exible appearance editing,\nthey also suffer from severe misalignment betw
 een geometry and high-frequency texture details. \nIn this work, we propos
 e a novel approach to boost the quality of generated 3D objects from the p
 erspective of Physics-Based Rendering (PBR) materials. \nBy analyzing the 
 components of PBR materials, \nwe choose to consider albedo, roughness, me
 talness, and normal bump in a single image to 3D object generation.\nFor a
 lbedo and normal-bump,\nwe leverage Stable Diffusion fine-tuned on synthet
 ic data to extract these values,\nwith novel usages of these fine-tuned mo
 dels to obtain 3D consistent albedo uv and normal-bump uv for generated ob
 jects.\nIn terms of roughness and metalness,\nwe adopt a semi-automatic pr
 ocess to provide room for interactive adjustment, which we believe is more
  practical.\nExtensive experiments demonstrate that our model is generally
  beneficial for various state-of-the-art generation methods, significantly
  boosting the quality and realism of their generated 3D objects, with natu
 ral relighting effects and substantially improved geometry.\n\nRegistratio
 n Category: Full Access, Full Access Supporter\n\nLanguage Format: English
  Language\n\nSession Chair: Valentin Deschaintre (Adobe Research)
URL:https://asia.siggraph.org/2024/program/?id=papers_967&sess=sess149
END:VEVENT
END:VCALENDAR