De Decker, AxelAxelDe DeckerSivashanmugama, SangarSangarSivashanmugamaWarrington, SteveSteveWarringtonDe Cock, JanJanDe CockVan Wallendael, GlennGlennVan Wallendael2026-04-132026-04-132025978-1-5106-9118-60277-786Xhttps://imec-publications.be/handle/20.500.12860/59057In video streaming and broadcasting, compression is essential to optimize bandwidth usage, but it often introduces visual artifacts that degrade the viewing experience. Maintaining high visual quality during encoding requires continuous perceptual quality assessment. This, in turn, demands methods that are both accurate and computationally efficient. Simple metrics such as PSNR and SSIM are widely used due to their speed, but they correlate poorly with human visual perception, while more accurate alternatives such as VMAF and MS-SSIM remain too computationally demanding for real-time use. We present predictive VMAF (pVMAF), a lightweight, in-loop quality metric that replicates VMAF scores during encoding. It relies on a set of bitstream statistics, low-complexity pixel attributes, and elementary quality metrics to estimate visual quality for each frame at a fraction of the cost of a full VMAF computation. Integrated into x264, x265, and SVT-AV1, we evaluate pVMAF in terms of implementation, prediction accuracy, and computational efficiency. Our results show that pVMAF closely tracks VMAF scores while introducing minimal computational overhead, making it a practical solution for real-time perceptual quality monitoring across multiple video codecs.engProceedings paper10.1117/12.3064449WOS:001680864100031TO-NOISE RATIO