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:20241204T165300
DTEND;TZID=Asia/Tokyo:20241204T170500
UID:siggraphasia_SIGGRAPH Asia 2024_sess122_papers_507@linklings.com
SUMMARY:StyleTex: Style Image-Guided Texture Generation for 3D Models
DESCRIPTION:Zhiyu Xie, Yuqing Zhang, Xiangjun Tang, Yiqian Wu, and Dehan C
 hen (State Key Laboratory of CAD&CG, Zhejiang University); Gongsheng Li (Z
 hejiang University); and Xiaogang Jin (State Key Laboratory of CAD&CG, Zhe
 jiang University)\n\nStyle-guided texture generation aims to generate a te
 xture that is harmonious with both the style of the reference image and th
 e geometry of the input mesh, given a reference style image and a 3D mesh 
 with its text description.  \nAlthough diffusion-based 3D texture generati
 on methods, such as distillation sampling, have numerous promising applica
 tions in stylized games and films, it requires addressing two challenges: 
 1) decouple style and content completely from the reference image for 3D m
 odels, and 2) align the generated texture with the color tone, style of th
 e reference image, and the given text prompt.\nTo this end, we introduce S
 tyleTex, an innovative diffusion-model-based framework for creating styliz
 ed textures for 3D models. Our key insight is to decouple style informatio
 n from the reference image while disregarding content in diffusion-based d
 istillation sampling.\nSpecifically, given a reference image, we first dec
 ompose its style feature from the image CLIP embedding by subtracting the 
 embedding's orthogonal projection in the direction of the content feature,
  which is represented by a text CLIP embedding. \nOur novel approach to di
 sentangling the reference image's style and content information allows us 
 to generate distinct style and content features. \nWe then inject the styl
 e feature into the cross-attention mechanism to incorporate it into the ge
 neration process, while utilizing the content feature as a negative prompt
  to further dissociate content information. \nFinally, we incorporate thes
 e strategies into StyleTex to obtain stylized textures. We utilize Interva
 l Score Matching to address over-smoothness and over-saturation, in combin
 ation with a geometry-aware ControlNet that ensures consistent geometry th
 roughout the generative process. The resulting textures generated by Style
 Tex retain the style of the reference image, while also aligning with the 
 text prompts and intrinsic details of the given 3D mesh.\nQuantitative and
  qualitative experiments show that our method outperforms existing baselin
 e methods by a significant margin.\n\nRegistration Category: Full Access, 
 Full Access Supporter\n\nLanguage Format: English Language\n\nSession Chai
 r: Minhyuk Sung (Korea Advanced Institute of Science and Technology (KAIST
 ))\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_507&sess=sess122
END:VEVENT
END:VCALENDAR