[MPlayer-users] LAVC-MPEG4 vs DIVX4 vs XVID??

[Matthias Wieser]

Am Mittwoch, 22. Oktober 2003 12:47 schrieb Tuukka Toivonen:

> >This command does surely not give the best looking results
>
> Impossible to say.

I woul say that the possibility is >90%

> I can not judge every option manually, but I am
> quite convinced that PSNR correlates with subjective quality quite well
> although not perfectly:

PSNR does not correlate with subjective quality quite well. For example 
block artifacts on walls or sky nearlly don't decrease psnr but they 
affect visual quality very much.
For example, see http://www.heise.de/ct/03/10/146/images/sarnoff.jpg
and http://www.jndmetrix.com/.

> >This command does surely not give the best results for jj2dog's (?)
> >testcase.
>
> If you know what options will, let me know =)

I don't know j2..'s testcase. So it's impossible to give a commandline for 
the "best results".

> Until somebody suggests
> better options (and also gives some justification why they're better),
> I would recommend mine.

Your testcase is very special: Noisy TV with super strong denoise3d.
J2d.. coult try qpel, which didn't increase the psnr of your video. But 
maybe it increases visual quality of his video file. You suggest 
predia=1:dia=1, but according to http://fatooh.org/remitests/part2.html 
predia and dia most times decrease video quality. And so on.