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 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241205T151900
DTEND;TZID=Asia/Tokyo:20241205T153100
UID:siggraphasia_SIGGRAPH Asia 2024_sess134_papers_816@linklings.com
SUMMARY:I2VEdit: First-Frame-Guided Video Editing via Image-to-Video Diffu
 sion Models
DESCRIPTION:Wenqi Ouyang (S-Lab for Advanced Intelligence, Nanyang Technol
 ogical University Singapore); Yi Dong (Nanyang Technological University (N
 TU)); Lei Yang and Jianlou Si (SenseTime); and Xingang Pan (S-Lab for Adva
 nced Intelligence, Nanyang Technological University Singapore)\n\nThe rema
 rkable generative capabilities of diffusion models have motivated extensiv
 e research in both image and video editing. Compared to video editing whic
 h faces additional challenges in the time dimension, image editing has wit
 nessed the development of more diverse, high-quality approaches and more c
 apable software like Photoshop. In light of this gap, we introduce a novel
  and generic solution that extends the applicability of image editing tool
 s to videos by propagating edits from a single frame to the entire video u
 sing a pre-trained image-to-video model. Our method, dubbed I2VEdit, adapt
 ively preserves the visual and motion integrity of the source video depend
 ing on the extent of the edits, effectively handling global edits, local e
 dits, and moderate shape changes, which existing methods cannot fully achi
 eve. At the core of our method are two main processes: Coarse Motion Extra
 ction to align basic motion patterns with the original video, and Appearan
 ce Refinement for precise adjustments using fine-grained attention matchin
 g. We also incorporate a skip-interval strategy to mitigate quality degrad
 ation from auto-regressive generation across multiple video clips. Experim
 ental results demonstrate our framework's superior performance in fine-gra
 ined video editing, proving its capability to produce high-quality, tempor
 ally consistent outputs.\n\nRegistration Category: Full Access, Full Acces
 s Supporter\n\nLanguage Format: English Language\n\nSession Chair: Nanxuan
  Zhao (Adobe Research)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_816&sess=sess134
END:VEVENT
END:VCALENDAR