VMAFi

VMAFi, commonly written VMAF or VMAFi in some sources, refers to the Video Multimethod Assessment Fusion metric used to predict perceptual video quality. It was developed by Netflix to provide a single, objective score that aligns with subjective human judgments of video quality. The term VMAFi is often used informally to denote the same metric or its inference-oriented implementations, but the core concept remains the same: a full-reference quality assessment that compares a distorted video to a high-quality reference.

VMAFi combines multiple quality features into a single score through a machine learning fusion model trained on human subjective ratings. It operates as a full-reference metric, requiring both a reference (original) video and a distorted version. The component features may reflect aspects such as structural similarity, detail preservation, information fidelity, and other perceptual cues. The resulting score typically ranges from 0 to 100, with higher values indicating better perceived quality. The approach aims to predict subjective quality across a wide range of content, codecs, resolutions, and bitrates.

In practice, VMAFi is implemented as an open-source library that can be used with media processing tools such as FFmpeg (via libvmaf) and provides interfaces in multiple programming languages. It is widely used in the video delivery industry to evaluate encoder settings, bitrate ladders, and compression strategies, and to compare codecs under controlled conditions. While useful and broadly adopted, VMAFi scores are best interpreted relative to consistent test conditions and datasets, and they do not guarantee identical human judgments across all viewing scenarios.