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:20260817T171537Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241204T171600
DTEND;TZID=Asia/Tokyo:20241204T172800
UID:siggraphasia_SIGGRAPH Asia 2024_sess122_papers_166@linklings.com
SUMMARY:Compositional Neural Textures
DESCRIPTION:Peihan Tu (University of Maryland, College Park); Li-Yi Wei (A
 dobe Research); and Matthias Zwicker (University of Maryland, College Park
 )\n\nTexture plays a vital role in enhancing visual richness in both real 
 photographs and computer-generated imagery. However, the process of editin
 g textures often involves laborious and repetitive manual adjustments of t
 extons, which are the recurring local patterns that characterize textures.
  This work introduces a fully unsupervised approach for representing textu
 res using a compositional neural model that captures individual textons. W
 e\nrepresent each texton as a 2D Gaussian function whose spatial support a
 pproximates\nits shape, and an associated feature that encodes its detaile
 d appearance. By modeling a texture as a discrete composition of Gaussian 
 textons, the representation offers both expressiveness and ease of editing
 . Textures can be edited by modifying the compositional Gaussians within t
 he latent space, and new textures can be efficiently synthesized by feedin
 g the modified Gaussians through a generator network in a feed-forward man
 ner. This approach enables a wide range of applications, including transfe
 rring appearance from an image texture to another image, diversifying text
 ures, texture interpolation, revealing/modifying texture variations, edit 
 propagation, texture animation, and direct texton manipulation. The propos
 ed\napproach contributes to advancing texture analysis, modeling, and edit
 ing techniques, and opens up new possibilities for creating visually appea
 ling images with controllable textures.\n\nRegistration Category: Full Acc
 ess, Full Access Supporter\n\nLanguage Format: English Language\n\nSession
  Chair: Minhyuk Sung (Korea Advanced Institute of Science and Technology (
 KAIST))\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_166&sess=sess122
END:VEVENT
END:VCALENDAR