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:20250110T023309Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241203T133400
DTEND;TZID=Asia/Tokyo:20241203T134600
UID:siggraphasia_SIGGRAPH Asia 2024_sess105_papers_697@linklings.com
SUMMARY:Customizing Text-to-Image Models with a Single Image Pair
DESCRIPTION:Technical Papers\n\nMaxwell Jones, Sheng-Yu Wang, and Nupur Ku
 mari (Carnegie Mellon University); David Bau (Northeastern University); an
 d Jun-Yan Zhu (Carnegie Mellon University)\n\nArt reinterpretation is the 
 practice of creating a variation of a reference work, making a paired artw
 ork that exhibits a distinct artistic style. We ask if such an image pair 
 can be used to customize a generative model to capture the demonstrated st
 ylistic difference. We propose Pair Customization, a new customization met
 hod that learns stylistic difference from a single image pair and then app
 lies the acquired style to the generation process. Unlike existing methods
  that learn to mimic a single concept from a collection of images, our met
 hod captures the stylistic difference between paired images. This allows u
 s to apply a stylistic change without overfitting to the specific image co
 ntent in the examples. To address this new task, we employ a joint optimiz
 ation method that explicitly separates the style and content into distinct
  LoRA weight spaces. We optimize these style and content weights to reprod
 uce the style and content images while encouraging their orthogonality. Du
 ring inference, we modify the diffusion process via a new style guidance b
 ased on our learned weights. Both qualitative and quantitative experiments
  show that our method can effectively learn style while avoiding overfitti
 ng to image content, highlighting the potential of modeling such stylistic
  differences from a single image pair.\n\nRegistration Category: Full Acce
 ss, Full Access Supporter\n\nLanguage Format: English Language\n\nSession 
 Chair: Kfir Aberman (Snap)
URL:https://asia.siggraph.org/2024/program/?id=papers_697&sess=sess105
END:VEVENT
END:VCALENDAR