Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion. Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules. |
27th March 2024, 14:25 | #9221 | Link | |
Registered User
Join Date: Dec 2013
Posts: 343
|
Quote:
|
|
27th March 2024, 23:49 | #9222 | Link | |
Moderator
Join Date: Jan 2006
Location: Portland, OR
Posts: 4,752
|
Quote:
Of course, VMAF isn't exactly a video quality metric - it is a video distortion metric. It doesn't try to say if video looks good or not, but how perceptually degraded from the source it will be. VMAF will rate a very accurate reencode of a very terrible quality source very high, and a pretty accurate reencode of a great looking source lower, even though viewers would say the latter looks much better. p1204.3 predicts how a viewer would relate the final quality delivered. As viewers don't have a source to compare to, that experience is fundamentally no-reference, so a no-reference metric can make sense. Quality and distortion metrics each have their place, and which is appropriate is based on what you're trying to accomplish. That said, I find p1204.3 to be more useful in encoder tuning than VMAF, even though a distortion metric would seem more appropriate. There has been some interesting early work done combining reference, bitstream, and baseband analysis for further subjective correlation improvements. The nice thing about Machine Learning is that you can throw in whatever base metrics you want to for training as long as you've got a good ground truth data set of subjective ratings. |
|
Thread Tools | Search this Thread |
Display Modes | |
|
|