[MPlayer-dev-eng] [RFC] preliminary x264 encoding help text

Loren Merritt lorenm at u.washington.edu
Sat Apr 30 02:20:17 CEST 2005


On Fri, 29 Apr 2005, Jeff Clagg wrote:

> I've found clips where the global PSNR optimal ip_factor is around 2.0,
> with an optimal pb_factor of 1.6 or so. The optimal settings were found
> by searching the option space for ip_factor in 0.1 intervals, then doing
> the same for pb_factor (this didn't take long since the clips were CIF
> and only a few minutes long). Does anyone else think these numbers seem
> ridiculously high?

I/P=2.0 does sound ridiculous.
B/P=1.6 doesn't sound bad a priori... XviD uses 1.5 - 2.0 depending on 
the quant (though they do some funnyness with mismatched quant and 
lambda). I have seen cases where large values of B/P increased global 
PSNR. But IMHO consistent quality looks better than greatly varying QP, 
even if PSNR says otherwise.

> Also, about chroma_qp_offset. In every case I've tried, -1 is the
> best value for global PSNR, but the difference isn't too big. Did
> the JVT make a mistake in choosing the characteristics of the chroma
> quantization table?

The weight that chroma plays in global PSNR is kinda arbitrary: It's half 
of luma because there are half as many chroma samples in YUV420. But if 
you pick some other subsampling ratio, the weight will be very different 
while an encoded video will look almost the same (assuming a well-chosen 
value of chroma_qp_offset). So it's possible that JVT decided that chroma 
was slightly less than half as important as luma.

However, I don't think that the discrepancy is part of the standard:
I've noticed that x264's luma PSNR may be (slightly) higher than Nero's,
while still giving 1 - 1.5 dB lower chroma. (And Nero always uses 
chroma_qp_offset=0). So it's probably something to do with our encoding 
decisions, but I don't know what.

Actually, I do know one thing that's different: the skip threshold for 
chroma blocks is higher than luma (both are using the JVT recommendation). 
But when I tried fiddling with it, the results were almost identical to 
those of chroma_qp_offset, so I didn't think it needed to be a separate 
option.

--Loren Merritt




More information about the MPlayer-dev-eng mailing list