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:20250110T023312Z LOCATION:Hall B5 (2)\, B Block\, Level 5 DTSTART;TZID=Asia/Tokyo:20241205T145600 DTEND;TZID=Asia/Tokyo:20241205T150800 UID:siggraphasia_SIGGRAPH Asia 2024_sess134_papers_561@linklings.com SUMMARY:Fashion-VDM: Video Diffusion Model for Virtual Try-On DESCRIPTION:Technical Papers\n\nJohanna Karras (Google Research, Universit y of Washington); Yingwei Li and Nan Liu (Google Research); Luyang Zhu (Go ogle Research, University of Washington); Innfarn Yoo, Andreas Lugmayr, an d Chris Lee (Google Research); and Ira Kemelmacher-Shlizerman (Google Rese arch, University of Washington)\n\nWe present Fashion-VDM, a video diffusi on model (VDM) for generating virtual try-on videos. Given an input garmen t image and person video, our method aims to generate a high-quality try-o n video of the person wearing the given garment, while preserving the pers on's identity and motion. Image-based virtual try-on has shown impressive results; however, existing video virtual try-on (VVT) methods are still la cking garment details and temporal consistency. To address these issues, w e propose a diffusion-based architecture for video virtual try-on, split c lassifier-free guidance for increased control over the conditioning inputs , and a progressive temporal training strategy for single-pass 64-frame, 5 12px video generation. We also demonstrate the effectiveness of joint imag e-video training for video try-on, especially when video data is limited. Our qualitative and quantitative experiments show that our approach sets t he new state-of-the-art for video virtual try-on.\n\nRegistration Category : Full Access, Full Access Supporter\n\nLanguage Format: English Language\ n\nSession Chair: Nanxuan Zhao (Adobe Research) URL:https://asia.siggraph.org/2024/program/?id=papers_561&sess=sess134 END:VEVENT END:VCALENDAR