BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:Asia/Shanghai BEGIN:STANDARD DTSTART:19910915T010000 TZOFFSETFROM:+0900 TZOFFSETTO:+0800 TZNAME:CST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20251115T161841Z UID:739F5BB3-8F36-4C6A-8178-BFA89523AEC1 DTSTART;TZID=Asia/Shanghai:20251112T103000 DTEND;TZID=Asia/Shanghai:20251112T115500 DESCRIPTION:The demand for high-quality and immersive visual content contin ues to outpace the capacity of current 5G and future 6G networks\, making compression an essential component of visual communication. Despite major advances in video coding over the past decades\, key challenges such as la tency\, energy efficiency\, scalability\, and robustness remain unresolved .\n\nThis talk will focus on one of these challenges: achieving effective compression at extremely low bitrates\, where traditional codecs fail to p reserve perceptual quality. Learning-based approaches enable substantial b andwidth reduction by exploiting the structure and semantics of visual con tent and can operate in a generative regime\, where visual data are recons tructed by conditioning a trained model on compact latent or semantic repr esentations. However\, efficiently navigating the rate–distortion–perc eption trade-off with these models remains a major open problem.\n\nI will illustrate these ideas through two examples: generative face video coding (GFVC)\, where realistic talking-face motion and texture can be synthesiz ed from compact transmitted features\, and generative 3D point cloud compr ession\, where compact embeddings are used to guide a diffusion-based reco nstruction. I will conclude by discussing how these concepts extend to sem antic and task-oriented video communication\, which generalizes traditiona l paradigms and opens new perspectives in this evolving field.\n\nBldg: Hu ang Danian Tea & Thinking House (UESTC)\, No. 555 Tianjiao Road\, Building 3C\, 8th Floor\, Zhongdian Sunshine Information Port\, Chengdu\, Sichuan\ , China LOCATION:Bldg: Huang Danian Tea & Thinking House (UESTC)\, No. 555 Tianjiao Road\, Building 3C\, 8th Floor\, Zhongdian Sunshine Information Port\, Ch engdu\, Sichuan\, China ORGANIZER:bing_li@uestc.edu.cn SEQUENCE:19 SUMMARY:Ultra-Low-Bitrate Compression of Visual Content with Generative AI: Toward Semantic Visual Communication URL;VALUE=URI:https://events.vtools.ieee.org/m/513496 X-ALT-DESC:Description: <br /><p class="MsoNormal"><span lang="EN-US">The d emand for high-quality and immersive visual content continues to outpace t he capacity of current 5G and future 6G networks\, making compression an e ssential component of visual communication. Despite major advances in vide o coding over the past decades\, key challenges such as latency\, energy e fficiency\, scalability\, and robustness remain unresolved.</span></p>\n<p class="MsoNormal"><span lang="EN-US">This talk will focus on one of these challenges: achieving effective compression at extremely low bitrates\, w here traditional codecs fail to preserve perceptual quality. Learning-base d approaches enable substantial bandwidth reduction by exploiting the stru cture and semantics of visual content and can operate in a generative regi me\, where visual data are reconstructed by conditioning a trained model o n compact latent or semantic representations. However\, efficiently naviga ting the rate&ndash\;distortion&ndash\;perception trade-off with these mod els remains a major open problem.</span></p>\n<p class="MsoNormal"><span l ang="EN-US">I will illustrate these ideas through two examples: generative face video coding (GFVC)\, where realistic talking-face motion and textur e can be synthesized from compact transmitted features\, and generative 3D point cloud compression\, where compact embeddings are used to guide a di ffusion-based reconstruction. I will conclude by discussing how these conc epts extend to semantic and task-oriented video communication\, which gene ralizes traditional paradigms and opens new perspectives in this evolving field.</span></p> END:VEVENT END:VCALENDAR