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:20250110T023312Z
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:Technical Papers\n\nZhiyu Xie, Yuqing Zhang, Xiangjun Tang, Yi
 qian Wu, and Dehan Chen (State Key Laboratory of CAD&CG, Zhejiang Universi
 ty); Gongsheng Li (Zhejiang University); and Xiaogang Jin (State Key Labor
 atory of CAD&CG, Zhejiang University)\n\nStyle-guided texture generation a
 ims to generate a texture that is harmonious with both the style of the re
 ference image and the geometry of the input mesh, given a reference style 
 image and a 3D mesh with its text description.  \nAlthough diffusion-based
  3D texture generation methods, such as distillation sampling, have numero
 us promising applications in stylized games and films, it requires address
 ing two challenges: 1) decouple style and content completely from the refe
 rence image for 3D models, and 2) align the generated texture with the col
 or tone, style of the reference image, and the given text prompt.\nTo this
  end, we introduce StyleTex, an innovative diffusion-model-based framework
  for creating stylized textures for 3D models. Our key insight is to decou
 ple style information from the reference image while disregarding content 
 in diffusion-based distillation sampling.\nSpecifically, given a reference
  image, we first decompose 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 disentangling the reference image's style and content in
 formation allows us to generate distinct style and content features. \nWe 
 then inject the style feature into the cross-attention mechanism to incorp
 orate it into the generation process, while utilizing the content feature 
 as a negative prompt to further dissociate content information. \nFinally,
  we incorporate these strategies into StyleTex to obtain stylized textures
 . We utilize Interval Score Matching to address over-smoothness and over-s
 aturation, in combination with a geometry-aware ControlNet that ensures co
 nsistent geometry throughout the generative process. The resulting texture
 s generated by StyleTex retain the style of the reference image, while als
 o aligning with the text prompts and intrinsic details of the given 3D mes
 h.\nQuantitative and qualitative experiments show that our method outperfo
 rms existing baseline methods by a significant margin.\n\nRegistration Cat
 egory: Full Access, Full Access Supporter\n\nLanguage Format: English Lang
 uage\n\nSession Chair: Minhyuk Sung (Korea Advanced Institute of Science a
 nd Technology (KAIST))
URL:https://asia.siggraph.org/2024/program/?id=papers_507&sess=sess122
END:VEVENT
END:VCALENDAR