[MPlayer-users] lavc vs. xvid (and improving lavc quality)

[Jason Tackaberry]

On Mon, 2004-06-07 at 18:19 +0200, Michael Niedermayer wrote:
> thats the idea, allthough that doesnt mean that it wont cause some other 
> artefacts, only actually using it will tell, theres also further finetuning 
> possible which should improve PSNR

Initial results are definitely good, but it's great to hear you have
some tweaks in mind.

I redid the blind test between xvid and lavc using 2-pass at 2200kbit
with NSSE.  Originally I picked xvid 5 out of 5 times, with 2 clips
being a close call.  This time I picked lavc 4 out of 5 times.

The clip where I preferred xvid over lavc was during a high action scene
where bitrate was a limiting factor, and qp began to drift upward for
the lavc version.  Likely if I'd ripped the whole movie, it would have
been able to give this scene a higher bitrate and I wouldn't have
noticed it a problem.

So it seems to be the case that, for higher quantizers, cmp=10 isn't as
good as the others, but for lower quantizers it looks superior (to me).
Perhaps this is stunningly obvious, but I thought I'd sound it out aloud
anyway just to see if it made sense. :)  It'd be nice if I could tune
things so that at higher quantizers, a different comparison function is
used.  Is this possible (or does it make sense)?

I played around a bit with the other cmp functions, and cmp=3 is a
pretty good improvement over cmp=0.  Right now, at least, I prefer cmp=3
with post-processing (spp,noise) over cmp=10 without post-processing.
But for those times when I want to circulate video with the expectation
that it won't have pp, NSSE is a significant improvement.  I get the
best of both worlds now.  I have very little to complain about. :)

Of course if I can get quality in cmp=10 that rivals cmp=3+pp, that
would be amazing. :)

Cheers,
Jason.