[MPlayer-users] MPlayer-1.0pre4 bug with divx4 multipass filter

[Vladimir Mosgalin]

On Sun, 12 Sep 2004, D Richard Felker III wrote:

DRFI>> Hmm.. Imho pp filter isn't slow. It's faster than mpeg4 decoding in
DRFI>> twisted cases, like: xvid + qpel decoded by xvid (it can't be decoded by
DRFI>> lavc),
DRFI>
DRFI>huh? lavc decodes any xvid file you throw at it. are you sure
DRFI>you're not using an _ancient_ libavcodec that's not getting updated
DRFI>when you cvs update? that would explain all your problems with
DRFI>lavc.....

Right now, I'm using dev-RPM-CVS-040904-03:12-3.3.2. Lavc was cvs up'ed
the same day, of course.

When trying to decode xvid-encoded content with lavc, there are random
distorted blocks appearing all over the image if qpel was used. If
"packed" xvid option was used, the situation is even worse. Personally,
I don't like qpel, so I don't suffer from this problem, but some fansub
groups encode with xvid and use qpel, so setting xvid as default decoder
is a good idea.

There are another problems when decoding with lavc. When running
"mplayer xvid-encoded-file.avi -vf noise=12uah", mplayer writes
"vf.c: have to REALLOCATE buffer memory :(" several times a second,
and often, but not always (in about 50% cases) segfaults on exit (in
uninit_vcodec), and then hangs. Interrupting doesn't help, only sigkill
does. Also, several upper lines get distorted.

Sure, -vf scale,noise hack fixes both problems, but -vfm xvid is much
cleaner solution!

DRFI>> which eats about 50% of my athlon 3200+ cpu on dvd resolution.
DRFI>
DRFI>pp filter doesn't work, in my experience. it still has blocks. you

It does. pp=ac, at least. I wonder why it isn't in pphelp page?
It works pretty nice. There are cases when spp is only solution, but
mostly even for low-quality image pp=ac is enough. It doesn't blur image
as much as spp, too.

DRFI>have to use spp (at least level 3 or 4 and preferably 5 or 6) which
DRFI>is very slow.

Levels 4/5/6 are nice, but the difference isn't all that great between 3
and higher levels. And I can see absolutely no difference between levels
5 and 6.

And spp blurs image, sometimes too much. I have to use unsharp=l3x3:1
after spp in a lot of cases...

Too bad that -autoq doesn't perform good with spp (it can even use level
0 sometimes, when cpu is powerful and resolution is low enough to use
level 4 all the time!)

DRFI>> Well, does it really matter if it is a "crap" or not when it comes to
DRFI>> codec comparison?
DRFI>no, but there could be serious differences between the codec needs of
DRFI>the two.

Sure, but the fact that ~maybe~ (I haven't tested much) lavc performs
better on anime isn't enough to say that lavc is always better, don't
you think? There are also regular movies from dvds and tv captures in
this world...

DRFI>and your eye can tell the difference between cutting off 4 pixels? :)

Umm... When comparing two pictures, yes. Also, the idea that I lost
something on the purpose will bug me if I will just let it be..
But my eye can't distinguish original & scaled picture (not with any sws
parameters, of course) after they both are upscaled to the full screen
size.

DRFI>i tried this, and saw blocks in two distinct places, with q=2:
DRFI>1) wherever the dvd already had blocks (yes dvds often have blocks!)

Yes, but mostly they have other kind of artifacts. And no, I'm not
talking about this case.

DRFI>2) in near-solid-black areas, where adjacent blocks only had the dc
DRFI>coefficient (solid color) and even 1 level difference between them was
DRFI>made visible by the insane gamma correction.

AFAIR, this is not the case either. Even if it effects a bit, it doesn't
explain why xvid performs much better (whis is the point of this
discussion).

-- 

Vladimir