[XviD-devel] Re: qpel smearing/noise problem

[Michael Niedermayer]

Hi

On Thursday 09 January 2003 16:41, Michael Niedermayer wrote:
> On Thursday 09 January 2003 16:13, Christoph Lampert wrote:
> > On Thu, 9 Jan 2003, Michael Niedermayer wrote:
> > > On Thursday 09 January 2003 13:56, James Hauxwell wrote:
> > > > In qpel mode you can have 8x8 and 16x16 blocks
> > > >
> > > > The process is
> > > > 8tap filter horizontal
> > > > Bilinear filter horizontal
> > > > 8tap filter vertical
> > > > Bilinear filter vertical
> > >
> > > damn, i forgot that, ffmpeg does what the standart says, not what the
> > > reference software does, so thats very likely the problem, only
> > > question is which is correct?
> >
> > The reference software is correct (well, in fact the standard has been
> > modified to be in compliance with the software. That's a clever way
> > of debugging: leave the software as it is, instead modify the specs. )
>
> ok, i believe u both, ill fix ffmpeg
ok, done

btw i allso reproduced the IDCT qpel difference problem 
(see http://www.mplayerhq.hu/~michael/qpel-test.avi) for the video, its 
encoded with qpel / simpleidct with different constant qpel vectors in 
different areas of the image, it cant be decoded without artifacts with a 
different IDCT, allthough its a unfair test as the constant pattern with 
constant vectors will lead to a catastrophic accumulation of errors (if a 
IDCT "error" ever happens in a block, then it will very likely happen over 
and over again for the same block ...)

[...]

Michael