[Ffmpeg-cvslog] CVS: ffmpeg/tests test.conf,1.10,1.11

[Michael Niedermayer]

Hi

On Sun, Feb 26, 2006 at 03:54:13PM +0100, Alex Beregszaszi wrote:
> Hi,
> 
> > > > >  Feed feed1.ffm
> > > > >  Format mpjpeg
> > > > > -Strict -1
> > > > 
> > > > Withouth strict jpeg doesnt even works here, that's why I added
> > > > that switch. The problem is the converison between yuv420 and
> > > > yuvj420 which is not done automatically.
> > > 
> > > So what about this? With that test.conf ffmpeg exists with error.
> > 
> > Indeed yes. And it seems that this also breaks jpg (despite them
> > having Strict -1 in their sections).
> > But with Strict -1 ffserver gives random results depending on
> > the arch you are using, so it's hardly usefull.

what about the bit-exact flag and (i)dct algo selection in ffserver?
later must be a portable C one not the auto select fastest ...


> > 
> > I can see 2 solutions:
> >  - the easy (and lame) one: remove the failing tests from regression
> >  tests. - the hard one: fix mjpeg (and possibly others) so that it
> >  gives
> >                  consistent results on every arch.
> 
> I dont know why would it give different results. With strict -1 the jpeg
> encoder will take a yuv420 input as yuvj420 input (the value ranges are
> different 16-235 vs 0-255 for luma) and do the encoding. Without it, no
> automatical yuv420->yuvj420 conversion happens, thus it bails out. The
> best would be adding automatic conversion.
> 
> Anyway, its interesting that it makes different results with yuv420.
> Michael, does strict -1 enables other features?

not that i know of

[...]

-- 
Michael