AV1 CBR padding function

[ckdt1105]

CBR Padding feature is seemed to fill the remaining bitrate with meaningless information so, bitrate on static screens doesn't drop much, so bitrate spikes are less and as a result, even if the buffer size is large (target bitrate X 2), can achieve almost constant bitrate.

On H264 and HEVC, the CBR Padding feature is works (probably) so bitrate spikes are very uncommon.

In AV1, the bitrate is reduced during static scenes this cause more bitrate spikes than H264 and HEVC.

In the current state, for strict CBR in AV1, the buffer size must be extremely lowered This means that there are losses in many places, such as lookahead quality and video quality.

[ckdt1105]

https://github.com/ma3uk/obs-qsv-onevpl/issues/5 is maybe my mistaking it for a bug, because the lack of CBR padding in AV1.

So, I think that the change of the default buffer size to 'TargetBitrate / 8' in 1.1.0.0_dev2 may not need.

And, it is possible that (Target bitrate/fps) x (LookAheadDepth - 1) = 'minimal buffersize' ex) (8000/60) x (40-1) = 5200, (8000/60) x (120-1) = 15866 so, current version is potential for problems with encoding modes like LA_EXT.

[ma3uk]

Thanks for the Issue. I am aware of this problem, all the conditions for CBR Padding to work with AV1 were met on my part, but it did not work, so a week ago I contacted Intel and gave them all the information, at the moment I am waiting for feedback from Intel

[ckdt1105]

@ma3uk Thanks for information. i hope intel can fix this problem.

[ckdt1105]

@ma3uk i found intel finally fix cbr problem https://github.com/obsproject/obs-studio/pull/10902

[ma3uk]

@ma3uk i found intel finally fix cbr problem obsproject/obs-studio#10902

Yes, I know this method, but if you look at the content of the PR that you threw off, you can see that it is designed for a new generation of GPUs starting with integrated Lunar Lake GPUs and discrete Battlemage GPUs. You can try to reproduce this fix by setting the WinBRCMaxAvgKbps parameters yourself to Bitrate 1.3, WinBRCSize to a value equal to the final FPS, and also enable CustomBufferSize and set it to Bitrate 2. You can test both with and without the fix enabled and post the results here, based on the results we will decide whether to enable this fix by default or leave everything as it is now, with the possibility of fine-tuning by the user. An update has recently been made to the Intel driver repository that adds a driver-level fix starting with Meteor Lake integrated GPUs.

[ckdt1105]

@ma3uk Thank you for your massive explain