BEGIN:VCALENDAR VERSION:2.0 PRODID:-//swoogo.com//NONSGML kigkonsult.se iCalcreator 2.41.90// CALSCALE:GREGORIAN UID:36626538-6638-4462-a161-366663386462 BEGIN:VEVENT UID:4aa3d833381ad5465c3b692f82bda2522812ec3c@swoogo.com DTSTAMP:20260817T142014Z DESCRIPTION:Video streaming has become a prevalent method for accessing mul timedia content\, where optimizing video quality while ensuring smooth pla yback is crucial. Traditional methods use a fixed bitrate ladder with pred etermined bitrate-resolution pairs\, which often fail to co-optimize bitra te and video quality. This mismatch results in either suboptimal video qua lity for a given bitrate or excessive bitrate for the desired quality. Thi s paper introduces a novel content-driven approach that predicts encoding parameters to achieve target perceptual video quality. We utilize the Vide o Multimethod Assessment Fusion (VMAF) metric\, which aligns closely with human perception and is derived from a machine learning model trained on e xtensive video content and subjective quality assessments. Our approach em ploys a deep learning model with a SlowFast Network Architecture\, analyzi ng video at two resolutions to predict a VMAF rate-distortion curve. This model uses higher-resolution frames for spatial information and more frequ ent lower-resolution frames for temporal details\, optimizing encoding eff iciency and perceptual quality. Our results demonstrate that this method r educes encoding attempts and cloud computing resources\, while maximizing video quality and encoding efficiency. DTSTART:20241024T163000Z DTEND:20241024T170000Z LAST-MODIFIED:20260817T142014Z LOCATION:Session Room 1 SEQUENCE:0 STATUS:CONFIRMED SUMMARY:Efficient Content driven Encoding Towards a Target Video Quality TRANSP:OPAQUE X-ALT-DESC;FMTTYPE=text/html:

Video streaming has become a prevalent meth od for accessing multimedia content\, where optimizing video quality while ensuring smooth playback is crucial. Traditional methods use a fixed bitr ate ladder with predetermined bitrate-resolution pairs\, which often fail to co-optimize bitrate and video quality. This mismatch results in either suboptimal video quality for a given bitrate or excessive bitrate for the desired quality. This paper introduces a novel content-driven approach tha t predicts encoding parameters to achieve target perceptual video quality. We utilize the Video Multimethod Assessment Fusion (VMAF) metric\, which aligns closely with human perception and is derived from a machine learnin g model trained on extensive video content and subjective quality assessme nts. Our approach employs a deep learning model with a SlowFast Network Ar chitecture\, analyzing video at two resolutions to predict a VMAF rate-dis tortion curve. This model uses higher-resolution frames for spatial inform ation and more frequent lower-resolution frames for temporal details\, opt imizing encoding efficiency and perceptual quality. Our results demonstrat e that this method reduces encoding attempts and cloud computing resources \, while maximizing video quality and encoding efficiency.

BEGIN:VALARM UID:38343963-3462-4263-b439-663462313364 ACTION:DISPLAY DESCRIPTION:Video streaming has become a prevalent method for accessing mul timedia content\, where optimizing video quality while ensuring smooth pla yback is crucial. Traditional methods use a fixed bitrate ladder with pred etermined bitrate-resolution pairs\, which often fail to co-optimize bitra te and video quality. This mismatch results in either suboptimal video qua lity for a given bitrate or excessive bitrate for the desired quality. Thi s paper introduces a novel content-driven approach that predicts encoding parameters to achieve target perceptual video quality. We utilize the Vide o Multimethod Assessment Fusion (VMAF) metric\, which aligns closely with human perception and is derived from a machine learning model trained on e xtensive video content and subjective quality assessments. Our approach em ploys a deep learning model with a SlowFast Network Architecture\, analyzi ng video at two resolutions to predict a VMAF rate-distortion curve. This model uses higher-resolution frames for spatial information and more frequ ent lower-resolution frames for temporal details\, optimizing encoding eff iciency and perceptual quality. Our results demonstrate that this method r educes encoding attempts and cloud computing resources\, while maximizing video quality and encoding efficiency. TRIGGER:-PT15M END:VALARM END:VEVENT END:VCALENDAR