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 B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241203T131100
DTEND;TZID=Asia/Tokyo:20241203T132300
UID:siggraphasia_SIGGRAPH Asia 2024_sess105_papers_824@linklings.com
SUMMARY:ReVersion: Diffusion-Based Relation Inversion from Images
DESCRIPTION:Ziqi Huang, Tianxing Wu, Yuming Jiang, Kelvin C.K. Chan, and Z
 iwei Liu (S-Lab for Advanced Intelligence, Nanyang Technological Universit
 y Singapore)\n\nDiffusion models gain increasing popularity for their gene
 rative capabilities. Recently, there have been surging needs to generate c
 ustomized images by inverting diffusion models from exemplar images, and e
 xisting inversion methods mainly focus on capturing object appearances (i.
 e., the "look"). However, how to invert object relations, another importan
 t pillar in the visual world, remains unexplored.\nIn this work, we propos
 e the Relation Inversion task, which aims to learn a specific relation (re
 presented as "relation prompt") from exemplar images. Specifically, we lea
 rn a relation prompt with a frozen pre-trained text-to-image diffusion mod
 el. The learned relation prompt can then be applied to generate relation-s
 pecific images with new objects, backgrounds, and styles. \n\nTo tackle th
 e Relation Inversion task, we propose the ReVersion Framework.\nSpecifical
 ly, we propose a novel "relation-steering contrastive learning" scheme to 
 steer the relation prompt towards relation-dense regions, and disentangle 
 it away from object appearances. \nWe further devise "relation-focal impor
 tance sampling" to emphasize high-level interactions over low-level appear
 ances (e.g., texture, color).\nTo comprehensively evaluate this new task, 
 we contribute the ReVersion Benchmark, which provides various exemplar ima
 ges with diverse relations. Extensive experiments validate the superiority
  of our approach over existing methods across a wide range of visual relat
 ions. Our proposed task and method could be good inspirations for future r
 esearch in various domains like generative inversion, few-shot learning, a
 nd visual relation detection.\n\nRegistration Category: Full Access, Full 
 Access Supporter\n\nLanguage Format: English Language\n\nSession Chair: Kf
 ir Aberman (Decart AI)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_824&sess=sess105
END:VEVENT
END:VCALENDAR