ACR — Absolute Category Rating — is the simplest standardised method for measuring perceived video quality with human viewers. Each viewer watches a single video clip in isolation and rates its quality on a 5-point scale: 5 = Excellent, 4 = Good, 3 = Fair, 2 = Poor, 1 = Bad. There's no reference clip to compare against; the viewer simply judges the clip on its own merits. Average the scores from a panel of viewers and you get a Mean Opinion Score (MOS) — the gold standard for "how does this look to a human?".
The method is defined in ITU-R Recommendation BT.500 alongside several alternatives (DSCQS, DCR, SSCQE), and ACR is the most commonly used in practice because of its simplicity: no reference to set up, no comparative judgements to make, just "rate what you see". Typical panels are 15–30 trained or untrained viewers, watched in controlled lab conditions with calibrated displays. The data feeds into codec benchmarks, perceptual metric training (VMAF was trained partly on ACR data), and product quality verification before launch.
For a product team, ACR is the ground truth that every objective metric is trying to approximate. PSNR, SSIM and VMAF all aim to predict ACR scores without needing humans in the loop — VMAF in particular was designed to correlate with ACR ratings as closely as possible. When you launch a new codec, a new bitrate ladder, or a new perceptual encoder, the final validation step usually involves running an ACR test with real viewers to confirm the change makes things better (or at least no worse) before rolling it out at scale. Vendors like Subjectify, Brightcove and several research-oriented services run commercial ACR panels for companies that need them.
