CAMBI, the Contrast-Aware Multiscale Banding Index, is a dedicated banding detector Netflix published at the Picture Coding Symposium in 2021 and open-sourced inside its VMAF repository. It exists because banding — the staircase of flat steps that appears in what should be a smooth sky, fade, or gradient — is the artifact the popular full-reference metrics are worst at: a visible band can differ from the true gradient by one code value, so PSNR can read about 48 dB and legacy VMAF 95 while the contours are plainly visible. CAMBI is a no-reference detector working frame by frame on a model of human vision, using the Contrast Sensitivity Function to weight each contour by how likely a viewer is to see it, so it flags the edges the eye catches, not harmless detail. Its score runs from 0 (no banding) up with severity, around 5 becoming slightly annoying and the worst values near 24. Visibility depends on viewing environment, and the 2026 VMAF v1 folds CAMBI in as a feature.
