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 (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:Yitong Wang (Fudan University, Shanghai Artificial Intelligenc
 e Laboratory); Xudong Xu (Shanghai Artificial Intelligence Laboratory); Li
  Ma (Scanline VFX Studio); Haoran Wang (Shanghai Jiaotong University); and
  Bo Dai (Shanghai Artificial Intelligence Laboratory)\n\nAutomatic 3D cont
 ent creation has gained increasing attention recently, due to its potentia
 l in various applications such as video games, film industry, and AR/VR. \
 nRecent advancements in diffusion models and multimodal models have notabl
 y improved the quality and efficiency of 3D object generation given a sing
 le RGB image. \nHowever, 3D objects generated even by state-of-the-art met
 hods are still unsatisfactory compared to human-created assets. \nConsider
 ing only textures instead of materials makes these methods encounter chall
 enges in photo-realistic rendering, relighting, and flexible appearance ed
 iting,\nthey also suffer from severe misalignment between geometry and hig
 h-frequency texture details. \nIn this work, we propose a novel approach t
 o boost the quality of generated 3D objects from the perspective of Physic
 s-Based Rendering (PBR) materials. \nBy analyzing the components of PBR ma
 terials, \nwe choose to consider albedo, roughness, metalness, and normal 
 bump in a single image to 3D object generation.\nFor albedo and normal-bum
 p,\nwe leverage Stable Diffusion fine-tuned on synthetic data to extract t
 hese values,\nwith novel usages of these fine-tuned models to obtain 3D co
 nsistent albedo uv and normal-bump uv for generated objects.\nIn terms of 
 roughness and metalness,\nwe adopt a semi-automatic process to provide roo
 m for interactive adjustment, which we believe is more practical.\nExtensi
 ve experiments demonstrate that our model is generally beneficial for vari
 ous state-of-the-art generation methods, significantly boosting the qualit
 y and realism of their generated 3D objects, with natural relighting effec
 ts and substantially improved geometry.\n\nRegistration Category: Full Acc
 ess, Full Access Supporter\n\nLanguage Format: English Language\n\nSession
  Chair: Valentin Deschaintre (Adobe Research)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_967&sess=sess149
END:VEVENT
END:VCALENDAR