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
fishgoesblub 3 hours ago [-]
I was hoping this would be a post announcing the death of VMAF from Netflix, alas.
r3trohack3r 3 hours ago [-]
Why?
fishgoesblub 2 hours ago [-]
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.
EA-3167 45 minutes ago [-]
At this point I've accepted that streaming is for the stuff that already doesn't look great (a lot of streaming content frankly), and for the rest there's physical media.
The difference is STARK.
shermantanktop 4 minutes ago [-]
The sensitivity to that difference is extremely varied. Customers buying TVs are attracted to ridiculously brightly colored videos, and go home to watch a murky movie with banding artifacts and say “I love my tv.” And they do.
I myself am not a golden eye, and I know some, and they can clap along to fragment boundaries during streaming.
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
The difference is STARK.
I myself am not a golden eye, and I know some, and they can clap along to fragment boundaries during streaming.