Re[2]: [XviD-devel] qpel - problems I found

[Michael Militzer]

Hi,

> > I don't think that this is correct either. The standard says that qpel
> > vectors are not simply halfed but rounded to the next halfpel value
using
> > this table:
>
> > fourth pixel position    0    1    2    3
> > resulting position        0    1    1    1
>
> Ok, I cancel my cvs update.
> This is because I really don't want to change good code with wrong.
> You're the bosses, tell me what should I do about it...
>
> > so:
> >     dx = (dx / 2) | (dx & 1);
> >     dy = (dy / 2) | (dy & 1);
>
> > should be ok then...
>
> No, I tried it. I just checked and it's ok indeed (just what the specs
> say) but ffdshow/divx5 quite obviously just divides by 2.

hm, strange. I had the opposite opinion: Last time I checked, DivX5 seemed
to also do some rounding, at least decoding DivX5 qpel clips with only
halfing qpel vectors for chroma blocks looked clearly wrong... Need to look
at it again.

> Another strange thing is that after the 'fix' filesize is smaller at
> fixed quant.
> Maybe mpeg4 found this out and changed the specs?

my specs are pretty recent (ISO second draft) and there's still this
rounding table (table 7-9) presented...

> As for the other fixes, I'm glad to hear that they are correct. They
> are all in motion_est.c so I can commit this file and I'll do it.

yes, please commit the fixed motion_est.c

bye,
Michael