WMV9 bidirectional frames performance

[noam_korem]

I've been playing around with several WMV encoders (including one from MS), and it seems like this standard exhibits low performance concerning B-frames.

While MPEG4's use of B frames usually reduce bitstream size by 10-20% (in relation to P-only streams), the WMV standard rarely shows any reduction at all (!).

The encoders I checked supported 1-7 consecutive B frames, 4MV, AC prediction, and all the knacks and gadgets available in the standard. It simply seems like WMV B frames are not cheap enough, while distancing the P-vops from one another demands a large amount of bits...

Can anyone comment on this? I'd like to know if this phenomenon is familiar, or am I doing something wrong?

NK.

[IvS]

What encoders have you tested?
Maybe the bitstream is already pretty optimized and therefore there's not much reduction? (Not sure, just a thought)

[noam_korem]

Used a microsoft encoder, amongst others.

This issue has appeared across several encoders, which leads me to think it is not encoder related.

My feeling is that there seems to be inherent issues with the Bidirectional frame in the wmv standard.
MPEG4 uses 1 bit for SKIPed macroblocks in P frames, and 0 bits for NOT_CODED macroblocks in B frames. This results in bit reduction when presenting the B frame feature into quiet scenes.
WMV uses 1 bit for SKIP in P frames, and 2 (!) bits for DIRECT SKIP in B frames.

NK

[IvS]

Interesting :)
It would be interesting to know whether this is really a format limitation or encoder limitation.
Also, I had no idea there are third party WMV9 encoders, front-ends - sure. If you could say which you've used I'd be interested in testing them myself.

[bond]

what other encoders? afaik m$ is the only one providing a vc-1 codec atm

[lazyn00b]

what other encoders? afaik m$ is the only one providing a vc-1 codec atm

Actually, there are tons of VC-1 implementations out there (Google is your friend!) - the bad news is that they are all insanely expensive. Take this one for example:

http://www.inlethd.com/products/fathom.htm

I think it costs about $25,000! Supposedly the quality of this software/hardware-assisted solution is much better than the "stock" encoder provided by Microsoft, and you can encode HD video in real-time.

[bond]

i assume these are all the same implementations: the one from m$

[IvS]

lazyn00b: Heh, yeah, that's a bit expensive.
I wanted to download the trial software to test if it indeed gives similar results to Microsoft's implementation, but got "Coming Soon" :).

[noam_korem]

My guess that this phenomenon will be present also in the VC-1 codec. So if anyone has got a VC1 encoder out there, I'd be happy to hear about its' performance as well

NK