Re[2]: [XviD-devel] xvid + bframes and mplayer/ffmpeg decoding

[Michael Niedermayer]

Hi

On Friday 16 August 2002 15:46, Christoph Lampert wrote:
> On Fri, 16 Aug 2002, Michael Niedermayer wrote:
> > Hi
> >
> > On Friday 16 August 2002 14:50, Christoph Lampert wrote:
> > > On Fri, 16 Aug 2002, Radoslaw Czyz wrote:
> > > > > Ouch! Is this in the standard? Where? Is this for B-frames, too?
> >
> > note 2 in 7.4.4.5
>
> Double ouch! I guess we don't do anything of 7.4.4.5. "Mismatch
> control"...
>
> This states that (for H263?) the sum of all unquantized coefficients
> must be calculated and if the result is _even_ then the last coefficient
> dct_codes[63] has to be changed by 1.
hmm, iirc that is only for mpeg-quantizers, its not obvious if note 2 applies 
to h263 too, but the problems occur with h263 quantizers so note 2  should 
applly to h263 ... i love clearly written standarts ;)

[...]
> And, of course the "DCT!=0 -> iDCT==0" thing... Maybe we should simply
> disable quant=1, I guess than the problem would be more or
> less solved... ;-)
i thought about the same for ffmpeg ...
btw, an alternative is to drop blocks with only a few +-1 coeffs and 0 for the 
rest (this seems to solve the issue with 1 test file and ffmpeg as encoder)

[...]

Michael