AVQT, Apple's Advanced Video Quality Tool, introduced at WWDC 2021, is a perceptual full-reference metric that estimates the quality of a compressed video on a 1-to-5 scale mirroring a Mean Opinion Score, reporting both per-frame and per-segment scores. Like VMAF and PSNR it takes the source and the compressed file, but two things set it apart. First, it is viewing-setup aware: it accepts display size, resolution, and viewing distance as explicit inputs and adjusts the predicted quality for how a real viewer in that setup would perceive the artifacts, so the same encode scores higher as the viewer sits farther away — a knob VMAF lacks. Second, it is fast and format-rich, built on AVFoundation and Metal (around 175 fps on 1080p) with native HDR10, HLG, and Dolby Vision handling. The catches are practical: AVQT runs only on macOS, which complicates Linux CI pipelines, and it is closed-source with no published model, so you cannot inspect or customize how a score was reached.
