Log in

View Full Version : The poor performance of B-frames bias


movmasty
7th January 2016, 23:37
Seems that to get the right quote of Bf you must use the optimal adaptive B frames,
but it costs a lot in time encoding, 25-30% more for a final pass,
and almost doubles the time for a fast 1st pass.

Looking for a workaround, setting fast adaptive, i tried to raise the bias, very poor results
then to raise max B together with bias, always poor results, few B more,
but really an higher Bias is raising the max number of consecutive B-frames, leaving the total almost equal.

The exact opposite of that an encoder would like, because the goal is to get more 1-2-3 consecutive B, lowering the 0 B points,turning 1 consecutive into 2, but to not have too many consecutive B.
With optimal 0 b frames is rare, with fast happens often, also if you raise the bias or max B frames.

There is any way to change this?
To modify the bias so that raising it will avoid no B frames in between or just one,
and wont raise 5-6 consecutive B.

And I am also addressing to the programmers of x265,
since those of x264 are by now in a relaxing mood(dont tell me that they are the same....)

movmasty
9th January 2016, 11:23
Disabling adaptive B works quite nice, more B frames, and a little adaptive too, because setting xB you get also some x-yB too

BUT, the number of I frames doubles, from 57 to 120 for 200 seconds of video,
to have an almost sane 85 i had to set an insane scenecut of 20,
and not sure if all these I are in the right place

Could this be fixed too? why no adaptive B increases I??