Netflix/vmaf

Perceptual video quality assessment based on multi-method fusion.

What it solves

VMAF (Video Multi-Method Assessment Fusion) solves the challenge of accurately measuring how humans perceive video quality. Unlike traditional mathematical metrics that may not align with human vision, VMAF provides a perceptual quality score that helps developers evaluate and optimize video compression and streaming performance.

How it works

The software uses a sophisticated algorithm to analyze video files and compare them against a reference source. It is available as a C library (libvmaf) for integration into software, a Python library for research and model training, and a command-line tool for production environments. It also supports integration with FFmpeg and includes tools for detecting specific artifacts like banding (via the CAMBI detector) and calculating other standard metrics like PSNR and SSIM.

Who it’s for

This project is for video engineers, researchers, and developers working on video streaming, codec evaluation, and quality assurance who need a reliable, industry-standard way to measure and improve video quality.

Highlights

  • Emmy-winning perceptual video quality assessment algorithm.
  • Includes support for HDR models and custom model training.
  • Provides a suite of additional metrics including PSNR, SSIM, MS-SSIM, and CIEDE2000.
  • Features CAMBI, a specialized detector for banding or contouring artifacts.
  • Highly flexible deployment options including C library, Python wrappers, Docker, and FFmpeg integration.

Related

  • Project
    slhck/ffmpeg-quality-metrics

    A command-line tool that calculates video quality metrics like PSNR, SSIM, VMAF, VIF, and MSAD by comparing a distorted video against a reference, outputting per-frame numbers and global statistics. It automates FFmpeg-based analysis and offers an optional interactive browser dashboard for visual inspection.

  • Project
    kornelski/dssim

    A tool that measures the visual similarity between images using an algorithm that mimics human perception, helping users benchmark image compression quality.

  • Project
    ldqk/ImageSearch

    A high-performance local tool for searching and deduplicating massive image and video collections using perceptual hashing algorithms.

  • Project
    Linfeng-Tang/SeAFusion

    A semantic-aware real-time network that fuses infrared and visible images to optimize them for high-level vision tasks like object detection and semantic segmentation.

  • Project
    LAION-AI/CLIP_benchmark

    A standardized evaluation framework for CLIP-like models to measure performance on zero-shot classification, retrieval, captioning, and linear probing across diverse datasets.