vmaf

Why we included this project

Anyone shipping compressed video has to be able to answer "how good does this look?" with something more concrete than a gut feeling, and VMAF is the metric Netflix built for exactly that job. It fuses several quality features into a single perceptual score that tracks human opinion more closely than raw pixel-error measures, and the same library also ships PSNR, SSIM, MS-SSIM and CIEDE2000, so the common metrics live in one place. The package is a C library (libvmaf) with a Python wrapper and command-line tooling, and it plugs straight into FFmpeg, which is why encoding teams use it to gate encodes or compare codecs and bitrate ladders without leaving their normal pipeline. It has become a reference metric for streaming quality work, and because the Python tools let you train and test your own VMAF models, teams can adapt the scoring to their own content instead of accepting a fixed model.