Netflix Releases libvmaf v3.0.0 and New VMAF v1 Models for Video Quality Assessment

libvmaf v3.0.0 Release

Netflix has released libvmaf v3.0.0, an update to its Emmy-winning perceptual video quality assessment algorithm. This new version incorporates several optimizations and bug fixes. It also fully removes APIs that were deprecated in the v2.0.0 release.

Introduction of VMAF v1 Models

Alongside the software update, Netflix has introduced a new set of VMAF models, designated as v1. These models provide updated capabilities for perceptual video quality assessment. The previous generation of models, v0, remains documented for reference.

VMAF Software Package

The VMAF software package includes a stand-alone C library, libvmaf, and a Python wrapping library. The Python library offers tools for users to train and test custom VMAF models. libvmaf also includes implementations of other metrics such as PSNR, PSNR-HVS, SSIM, MS-SSIM, and CIEDE2000.

CAMBI Integration

Netflix previously open-sourced CAMBI (Contrast Aware Multiscale Banding Index), a detector for banding artifacts. Recent updates to CAMBI included adding a full_ref input parameter for full-reference metric usage and a max_log_contrast parameter to capture banding with higher contrasts. CAMBI also saw speed improvements, such as a 4.5x speedup for 4K content.