Netflix/vmaf commit 47d0655 changes API of compute_vmaf again

[LigH]

Related to ffmpeg trac ticket #6921 – but new:

​vmaf commit 47d0655 adds two more parameters to compute_vmaf_score(..., int n_thread, int n_subsample)

The whole API may have to be treated as unstable.

[LigH]

compute_vmaf() in wrapper/src/libvmaf.cpp and wrapper/src/libvmaf.h

[LigH]

I don't have the experience to create a correct diff file, so just ​quoting a suggestion by 'timothymowens':

Heads-up, I was able to get it to build by replacing this on line 169 of build\ffmpeg-git\libavfilter\vf_libvmaf.c:
s->ms_ssim, s->pool);
with this:
s->ms_ssim, s->pool, 0, 1);

Not a permanent solution, but will get me by until this issue is resolved correctly by ffmpeg/vmaf teams.

[Carl Eugen Hoyos]

The most important thing about this ticket is probably that we’ll have to remove this dependency if no stable api is provided, see also the discussion about the Turing library.

[LigH]

And more changes to the VMAF API:

• v1.3.7 {​1b0f6ef}: added bool conf_interval
• v1.3.8 {​f5ea345}: changed to int enable_conf_interval

Does 'conf' mean "confusion"?

​MABS clamps to a last stable libvmaf revision before all these changes now.

[Carl Eugen Hoyos]

Maybe fixed in 87cc7e8d4ef8fa643d8d4822525b9c95cc9e7307