Quote:
Originally Posted by savage747
I dusted off my own automated codec testing setup so I can track AV1-development. Of course, objective metrics do not replace subjective testing, but this should still be okay to get some very rough estimates.
|
It can be VERY rough. It would be pretty typical for a new psychovisual feature to result in improved quality AND reduced scores with objective metrics. That is pretty much the defining feature of a psychovisual optimization
.
The following graph takes about 45 minutes to generate on my Ryzen 2700 (8 cores, 16 threads). To keep things sane, I'm using --cpu-used=4, which is a somwhat "fast" setting.
Quote:
The modern codecs may perform even better on HD resolutions.
|
HEVC certainly has a bigger advantage over H.264 at UHD resolutions.
It's hard to say with AV1 since the encoders are so slow that there isn't a substantial corpus of high-quality HD AV1 encodes to evaluate yet. In theory it should also scale well, but today's encoders obviously haven't been able to get a lot of resolution-tuned psychovisual optimization yet.
A whole lot of AV1's discussed capabilities are more informed speculation than anything based on real-world demonstrations of real-world scenarios.