What is VMAF?
VMAF, short for Video Multimethod Assessment Fusion, is a full-reference video quality metric introduced by Netflix alongside the University of Southern California.
The primary responsibility of VMAF is to predict video quality based on a reference and distorted video sequence. Netflix and other major platforms use this key metric to compare different video codecs, encoding configurations, encoders, and so on.
Key Components of VMAF
- Visual Information Fidelity (VIF): Used for factoring in information fidelity loss on four spatial scales.
- Detail Loss Metric (DLM): Used for measuring quality degradation, loss of details, and noticeable impairment in video streams.
- Mean Co-Located Pixel Difference (MCPD): Used for measuring the temporal difference between the frames on luminance components.
Real-Time Performance
- Regarding prediction precision, VMAF has been shown to outperform other primary image and video quality metrics, including SSIM, PSNR-HVS, and VQM-VFD based on performance on three to four data sets.
- As per a 2017 set, engineers from RealNetworks determined that VMAF has good reproducibility as per Netflix’s performance findings.
- As per the MSU Video Quality Metrics benchmark test, VMAF outperformed all the other metrics (including NEG), which were tested based on all compression standards (H.265, VP9, AV1, VVC).
- There are ways to improve VMAF scores artificially without improving the perceived video quality by applying various functions before distorting the video.
Benefits of Using VMAF
- VMAF outperforms other metrics, including the likes of PSNR and SSIM, based on accuracy and complexity.
- VMAF uses several traditional metrics that measure perceived video quality in a data-driven manner to show practical and more accurate results.
- VMAF is an open-source package, easily accessible and widely available.
- VMAF has reliable support, uses a high-quality codebase, and has functional tools and libraries for enhanced results.