Why aren't B-frame quantizers dynamic?

[drcl]

Why are they set according to a I/P frame ratio and offset calculation?

[DarkZell666]

Tha'ts a joly good question, for which an answer would be interesting indeed :)

[Didée]

The question "Why aren't B-frame quantizers dynamic?" is not quite correct. B-frame quantizers are dynamic. Just as dynamic as P-frame quantizers are. :)

[Sharktooth]

I think he meant "adaptive" or something like that...

[chilledoutuk]

i really think he means why is the quant always associated to the p frames it refers to as oposed to being set independantly using some rate controll alglorithm.

[foxyshadis]

But it's not like the b-frame quant is based on the last p-frame. Afaik, the codec calculates what the quant would be if it were a p-frame, and if it's an i lowers it, if it's a b it applies the ratio and offset. The quantizers of b-frames will fluctuate just as much as the other frame types, in a very similar fashion to the way the p-frames would if the stream was made again without b-frames.

[akupenguin]

The short answer is: because it works.

No one has found a fast algorithm that gives better results than a constant offset/ratio. There's an experimental brute-force decision in x264 (see "rdrc"), but it's dead slow and not well integrated with the rest of ratecontrol. And sure enough, the B-frame offset chosen by brute-force doesn't vary all that much; most of its benefits come from selecting the base I- and P-frame qps.

[drcl]

in XVID encoder tooltips....

BVOP Quant = AVG(past_VOP_quant, future_VOP_quant) * quant_ratio + quant_offset