[MPlayer-dev-eng] [PATCH] H.264 encoding in mencoder

[Loren Merritt]

On Tue, 31 Aug 2004, Romain Dolbeau wrote:

> With this patch, works for me. Seems decently fast with
> a (very slightly) higher PSNR than both xvid and lavc

> I take it back. With carefully chosen parameters, it
> seems lavc still manage to best the others, on Y
> and overall. But x264 is almost the same quality and
> apprently faster....

My results:
At constant QP, x264 has about .25  higher PSNR than anything I can
get from lavc (equivalently, 5% smaller at the same PSNR).
When comparing 2 pass ratecontrol, lavc with carefully chosen parameters
(half the speed of x264) can produce about .15 higher psnr than x264 (i.e.
3% smaller).

> (assuming the Y/U/V PSNR can be compared...).

x264's "Global" PSNR is comparable to lavc's "All".
If you use  -x264encopts log=3  then it prints stats for each frame that
are comparable to lavc's psnr.log

But x264's "Mean Y,U,V" and "Avg" use an arithmetic mean of the frame-wise
PSNRs, which is comparable to JM but not to lavc or xvid.
(And, BTW, this mean is theoretically unsound: a few black frames with no
error at all (PSNR=infinity but capped at 100) will completely throw off
the mean.)

--Loren Merritt