Install libvmaf with Homebrew

Project links

• Homepage

Commands and aliases

• vmaf

Project history

Netflix developed VMAF to better predict viewer-perceived streaming video quality than simple signal metrics such as PSNR. Netflix's own reference list traces VMAF through earlier academic work on fusion-based video quality assessment and notes that the project was open-sourced on GitHub in June 2016.

The first public implementation combined a C feature-extraction core with Python control code for experimentation. By October 2018 Netflix described the addition of a standalone executable, AVX acceleration in December 2016, frame-level multithreading in June 2018, and the packaging of the implementation as libvmaf with help from the FFmpeg community.

Adoption history

VMAF became part of practical codec evaluation because Netflix used it for production encoding decisions, Dynamic Optimizer workflows, and A/B experimentation around video quality. The GitHub README also documents FFmpeg support through configure --enable-libvmaf and AOM common test-condition usage, which made the library a package-manager staple for video-toolchain builds.

How it is used

Package users install libvmaf to run the vmaf command, link the C library into quality-analysis tools, train or test models with the Python layer, or enable FFmpeg's libvmaf filter. It is especially useful when comparing encoders, bitrates, resolutions, or codec changes on a perceptual scale.

Why package nerds care

For package nerds, libvmaf is the quality-metric dependency that turns FFmpeg from a transcoder into a measurable codec lab. Its presence in package managers signals a mature media stack: SIMD-heavy native code, model files, Python tooling, and a bridge between academic video-quality research and production streaming workflows.

Timeline

• 2014-2015: Academic work on fusion-based and ensemble-learning video quality metrics appears in the VMAF reference list.
• 2016-06: Netflix open-sources VMAF on GitHub.
• 2016-12: Netflix adds AVX optimization to VMAF's convolution work.
• 2018-06: Frame-level multithreading and frame skipping make practical high-resolution runs faster.
• 2018-10-25: Netflix describes libvmaf packaging and FFmpeg filter adoption in a VMAF project update.
• 2020-12-03: libvmaf v2.0.0 is released with a new optimized implementation and API.
• 2023-12-07: libvmaf v3.0.0 removes APIs deprecated in v2.0.0.
• 2026-06: Netflix announces VMAF model v1 documentation and model-selection guidance.

Related projects

• VMAF is closely related to FFmpeg, AOM codec test conditions, Netflix's Dynamic Optimizer work, CAMBI for banding detection, and traditional objective metrics such as PSNR, SSIM, MS-SSIM, and PSNR-HVS that are also exposed by the package.

Sources

• https://github.com/Netflix/vmaf
• https://github.com/Netflix/vmaf/blob/master/resource/doc/references.md
• https://netflixtechblog.com/vmaf-the-journey-continues-44b51ee9ed12
• https://research.netflix.com/publication/toward-a-better-quality-metric-for-the-video-community