Video Multimethod Assessment Fusion

Video Multimethod Assessment Fusion, known as VMAF, is an objective full-reference video quality metric developed by Netflix in cooperation with the University of Southern California, Nantes Universite and the University of Texas at Austin. It takes a reference video and a distorted version and predicts how a viewer would rate the quality of the distorted one, so it can be used to compare codecs, encoders, encoder settings or transmission variants. The metric fuses several existing image quality measures with a support vector machine regression into one score per frame between 0 and 100, where 100 means quality identical to the reference, and then averages the frame scores over the whole sequence. The original 2016 version fused visual information fidelity, a detail loss metric and a temporal difference measure. Version 1 uses the detail loss metric, a banding index, a chroma feature and the temporal measure, and adjusts for viewing distance so one model can handle 4K. The reference implementation is free software, and the work received a Technology and Engineering Emmy Award in 2021.