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:20241205T172800 DTEND;TZID=Asia/Tokyo:20241205T174000 UID:siggraphasia_SIGGRAPH Asia 2024_sess137_papers_920@linklings.com SUMMARY:MV2MV: Multi-View Image Translation via View-Consistent Diffusion Models DESCRIPTION:Technical Papers\n\nYoucheng Cai, Runshi Li, and Ligang Liu (U niversity of Science and Technology of China)\n\nImage translation has var ious applications in computer graphics and computer vision, aiming to tran sfer images from one domain to another. Thanks to the excellent generation capability of diffusion models, recent single-view image translation meth ods achieve realistic results. However, directly applying diffusion models for multi-view image translation remains challenging for two major obstac les: the need for paired training data and the limited view consistency. T o overcome the obstacles, we present a unified multi-view image to multi-v iew image translation framework based on diffusion models, called MV2MV. F irstly, we propose a novel self-supervised training strategy that exploits the success of off-the-shelf single-view image translators and the 3D Gau ssian Splatting (3DGS) technique to generate pseudo ground truths as super visory signals, leading to enhanced consistency and fine details. Addition ally, we propose a latent multi-view consistency block, which utilizes the latent-3DGS as the underlying 3D representation to facilitate information exchange across multi-view images and inject 3D prior into the diffusion model to enforce consistency. Finally, our approach simultaneously optimiz es the diffusion model and 3DGS to achieve a better trade-off between cons istency and realism. Extensive experiments across various translation task s demonstrate that MV2MV outperforms task-specific specialists in both qua ntitative and qualitative.\n\nRegistration Category: Full Access, Full Acc ess Supporter\n\nLanguage Format: English Language\n\nSession Chair: Micha el Rubinstein (Google) URL:https://asia.siggraph.org/2024/program/?id=papers_920&sess=sess137 END:VEVENT END:VCALENDAR