VMAF is super cool, although not the end-all-be-all of video quality measurement for sure. I'm currently evaluating switching over to the new v1.0 models.
The (unsolicited) slop PRs on that repository are a sight to behold, though. The same slopper posted an aggressive response to my PR: https://github.com/Netflix/vmaf/pull/1586
It's a flawed, and unreliable metric. Other metrics like SSIMULACRA2, CVVDP, or Butteraugli are more reliable/accurate and should be used if one cares about quality.
Blog post about the latest version: https://netflixtechblog.com/vmaf-v1-good-is-not-good-enough-...
VMAF is super cool, although not the end-all-be-all of video quality measurement for sure. I'm currently evaluating switching over to the new v1.0 models.
The (unsolicited) slop PRs on that repository are a sight to behold, though. The same slopper posted an aggressive response to my PR: https://github.com/Netflix/vmaf/pull/1586
I was hoping this would be a post announcing the death of VMAF from Netflix, alas.
Why?
It's a flawed, and unreliable metric. Other metrics like SSIMULACRA2, CVVDP, or Butteraugli are more reliable/accurate and should be used if one cares about quality.