B-frame bias needs to be much stronger?

[Zep]

I use 3 b-frames as my default on most everything I encode but I have noticed even with b-frame bias maxed to 100 in megui the most I can get is about 1.2 b frames to every p frame. Is this normal? Is there a way to get more b-frames? When I bumped the number allowed in a row it made no difference.

what I was hoping to get was 2 B for every P :)


thoughts?

[Dark Shikari]

More B-frames isn't necessarily optimal. x264's B-frame decision, on the other hand, isn't that great.

If you find that you get better efficiency for a source with more frames, you can try a more bruteforce method like --bframes 3 --no-b-adapt --b-pyramid, but its not really recommended.

Generally unless you can prove otherwise, its best to trust x264's B-frame decision.

[audyovydeo]

I use 3 b-frames as my default on most everything I encode but I have noticed even with b-frame bias maxed to 100 in megui the most I can get is about 1.2 b frames to every p frame. Is this normal? Is there a way to get more b-frames? When I bumped the number allowed in a row it made no difference.

what I was hoping to get was 2 B for every P :)


thoughts?

some time ago I played around with b-bias (setting bframes = 16) to see what would give. I deleted my results now but I remember the highest number of consecutive bframes I managed to get was 7 (once on a 2000-frame clip) with typically 2-to-5 bframe "lumps".

but this was academic, for how do you know that you're "improving" on x264's decision-making ? For each & every source ?

So I stand by Darkie's statement, & let x264 decide.

cheers
audyovydeo

[Zep]

More B-frames isn't necessarily optimal. x264's B-frame decision, on the other hand, isn't that great.

If you find that you get better efficiency for a source with more frames, you can try a more bruteforce method like --bframes 3 --no-b-adapt --b-pyramid, but its not really recommended.

Generally unless you can prove otherwise, its best to trust x264's B-frame decision.

well all I can say about that is the b-frames look as good as the p frames in motion and in my test example above they were 1/5 the size of p frames. I would think you would save more bits than lose to p frame size increase. Anyway, if b frames were only 1/2 the size of p frames I would doubt it would be a positive over all gain. I just wish I could TEST to make sure, which of course that was what I was trying to do when I ran into the b frame limit wall :)


thx

[Zep]

some time ago I played around with b-bias (setting bframes = 16) to see what would give. I deleted my results now but I remember the highest number of consecutive bframes I managed to get was 7 (once on a 2000-frame clip) with typically 2-to-5 bframe "lumps".


I have never seen more than 5 but I get 2+ p frames in a row too which has me wondering what could possibly be in the video that x264 core thinks no b frame at all is the best way to go...


but this was academic, for how do you know that you're "improving" on x264's decision-making ? For each & every source ?

So I stand by Darkie's statement, & let x264 decide.

cheers
audyovydeo

the Q average is lower? it looks better? more detail less blocking etc... easy to tell when one way is better than another. if you can't tell then it is not enough to worry about :)

I doubt x264 is as good as a human testing settings against all types of sources like my football 60 FPS encodes. Xvid more b frames help a lot with certain sources like 60 FPS sporting events. The difference between 1 b frame and 3 ratio to p frames is night and day on over all quality. The less change between frames the more valuable b frames are in Xvid. This is what I wanted to test with x264 :D