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

[Loren Merritt]

On Mon, 13 Sep 2004, Vladimir Mosgalin wrote:
> On Sun, 12 Sep 2004, D Richard Felker III wrote:
>
>>> 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.
>>
>> this has nothing to do with xvid/lavc...and in fact it should never
>
> But it has.
>
>> happen. if you see a crazy bug like this, please upload samples and
>> make a bug report! obviously no one else has experienced this problem,
>> or at least none of the developers.
>
> OK. You can get one yourself, by encoding any file with xvid; I'll post
> options that will lead to this when encoding the file I'll upload, to be
> sure.

I've never encountered that bug in the wild, but...

On Sun, Sep 12, 2004, Vladimir Mosgalin wrote:
> I always try to keep original dvd resolution, without rescaling, to keep
> all the details. When width and height aren't multiplies of 8 (16 is too
> harsh) and I don't want to crop too much, I scale, say, from 708 to 704
> pixels - eye can't tell the difference.

There's your problem. After a little fiddling, I can reproduce your
bug with:
video resolution not divisible by 16,
encoded with xvid, (ver: 1.0.2)
decoded with lavc, (ver: cvs 20040910)
and a -vf filter chain beginning with spp, unsharp, noise, or delogo
(don't ask me what's special about those filters)


BTW, multiples of 16 (not just 8) really are a good idea, and are very 
cheap compared to everything else you could do to improve video quality: 
At a given PSNR (and I see no reason why this comparison shouldn't be 
adequately measured by PSNR), A mpeg4 video with both dimensions 
non-mod16 takes 5-10% higher bitrate than one cropped, downscaled, or
even upscaled to the nearest mod16.
The penalty is about half that if only one dimension is non-mod16.
Even with 4mv and a resolution divisible by 8, you don't get much of 
that back (though it depends on the content).
Maybe I'll get around to posting graphs/writeups of my mod16 benchmarks 
sometime...

--Loren Merritt