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:20260817T171533Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241204T105600
DTEND;TZID=Asia/Tokyo:20241204T110800
UID:siggraphasia_SIGGRAPH Asia 2024_sess114_papers_716@linklings.com
SUMMARY:SGEdit: Bridging LLM with Text2Image Generative Model for Scene Gr
 aph-based Image Editing
DESCRIPTION:Zhiyuan Zhang (City University of Hong Kong), DongDong Chen (M
 icrosoft GenAI), and Jing Liao (City University of Hong Kong)\n\nScene gra
 phs offer a structured, hierarchical representation of images, with nodes 
 and edges symbolizing objects and the relationships among them. It can ser
 ve as a natural interface for image editing, dramatically improving precis
 ion and flexibility. Leveraging this benefit, we introduce a new framework
  that integrates large language model (LLM) with Text2Image generative mod
 el for scene graph-based image editing. This integration enables precise m
 odifications at the object level and creative recomposition of scenes with
 out compromising overall image integrity. Our approach involves two primar
 y stages: 1) Utilizing a LLM-driven scene parser, we construct an image's 
 scene graph, capturing key objects and their interrelationships, as well a
 s parsing fine-grained attributes such as object masks and descriptions. T
 hese annotations facilitate concept learning with a fine-tuned diffusion m
 odel, representing each object with an optimized token and detailed descri
 ption prompt. 2) During the image editing phase, a LLM editing controller 
 guides the edits towards specific areas. These edits are then implemented 
 by an attention-modulated diffusion editor, utilizing the fine-tuned model
  to perform object additions, deletions, replacements, and adjustments. Th
 rough extensive experiments, we demonstrate that our framework significant
 ly outperforms existing image editing methods in terms of editing precisio
 n and scene aesthetics. Our code will be made publicly available.\n\nRegis
 tration Category: Full Access, Full Access Supporter\n\nLanguage Format: E
 nglish Language\n\nSession Chair: Kai Wang (Simon Fraser University, Shang
 haiTech University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_716&sess=sess114
END:VEVENT
END:VCALENDAR