Extension of rate control to support GOP size of 32 (JVET-T0062).

[liufeiyang]

Rate control in current VTM does not support random access configuration of GOP size equals to 32.

[zhou]

Before merging the code for this fix, it would be great to test how the updated rate control algorithm performs w.r.t. the fixed QP coding. One could run the CTC first, then code the same sequences again with the rate control on and the resulting bit-rates of the fixed QP coding as the target bit-rates, then compute the BD-rate difference between the two.

[liufeiyang]

Replying to zhou:

Before merging the code for this fix, it would be great to test how the updated rate control algorithm performs w.r.t. the fixed QP coding. One could run the CTC first, then code the same sequences again with the rate control on and the resulting bit-rates of the fixed QP coding as the target bit-rates, then compute the BD-rate difference between the two.

Thanks for your kindly advice. This ticket is created according to the advices from software coordinator. We have conducted the experiments on common test sequences and the results are shown in the proposal JVET-T0062.

[fbossen]

Minhua: I expect the following to be triggered if you try to use rate control with a GOP size of 32:

msg( WARNING, "\n hierarchical bit allocation is not support for the specified coding structure currently.\n" );

because you have things like

else if ( GOPSize == 8 && !isLowdelay )
else if ( GOPSize == 16 && !isLowdelay )

in the code.

The purpose of the proposal is to handle the case where GOPSize == 32.

The test you are suggesting is something one could do after fixing this, not before.

The proposed code changes are here:
​https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/-/merge_requests/1909