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:20260817T171531Z
LOCATION:Hall B5 (1)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241204T145900
DTEND;TZID=Asia/Tokyo:20241204T151300
UID:siggraphasia_SIGGRAPH Asia 2024_sess118_papers_671@linklings.com
SUMMARY:FabricDiffusion: High-Fidelity Texture Transfer for 3D Garments Ge
 neration from In-The-Wild Images
DESCRIPTION:Cheng Zhang (Carnegie Mellon University, Texas A&M University)
 ; Yuanhao Wang and Francisco Vicente (Carnegie Mellon University); Chengle
 i Wu, Jinlong Yang, and Thabo Beeler (Google Inc.); and Fernando De la Tor
 re (Carnegie Mellon University)\n\nWe introduce FabricDiffusion, a method 
 for transferring fabric textures from a single clothing image to 3D garmen
 ts of arbitrary shapes. Existing approaches typically synthesize textures 
 on the garment surface through 2D-to-3D texture mapping or depth-aware inp
 ainting via generative models. Unfortunately, these methods often struggle
  to capture and preserve texture details, particularly due to challenging 
 occlusions, distortions, or poses in the input image. Inspired by the obse
 rvation that in the fashion industry, most garments are constructed by sti
 tching sewing patterns with flat, repeatable textures, we cast the task of
  clothing texture transfer as extracting distortion-free, tileable texture
  materials that are subsequently mapped onto the UV space of the garment. 
 Building upon this insight, we train a denoising diffusion model with a la
 rge-scale synthetic dataset to rectify distortions in the input texture im
 age. This process yields a flat texture map that enables a tight coupling 
 with existing Physically-Based Rendering (PBR) material generation pipelin
 es, allowing for realistic relighting of the garment under various lightin
 g conditions. We show that FabricDiffusion can transfer various features f
 rom a single clothing image including texture patterns, material propertie
 s, and detailed prints and logos. Extensive experiments demonstrate that o
 ur model significantly outperforms state-to-the-art methods on both synthe
 tic data and real-world, in-the-wild clothing images while generalizing to
  unseen textures and garment shapes.\n\nRegistration Category: Full Access
 , Full Access Supporter\n\nLanguage Format: English Language\n\nSession Ch
 air: Meng Zhang (Nanjing University of Science and Technology)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_671&sess=sess118
END:VEVENT
END:VCALENDAR