[XviD-devel] Cartoon mode

Christoph Lampert chl at math.uni-bonn.de
Fri Apr 18 14:55:05 CEST 2003


Ooops, this was meant for Micahel... please ignore
and don't you all try to download!!! ;-)

gruel

On Fri, 18 Apr 2003, Christoph Lampert wrote:
> On Thu, 17 Apr 2003, Michael Militzer wrote:
> 
> > Hi,
> > 
> > Quoting Christoph Lampert <chl at math.uni-bonn.de>:
> > 
> > > 
> > > Hi,
> > > 
> > > with all the new and upcoming option, we should maybe bundle
> > > our knowledge into some "Presets" for the GUI. Instead of 
> > > suggestions like "quarterpel might help, but not always, depends on the
> > > input material" (which is absolutely right), we might come up with 
> > > good presets for the most frequently encoded material. 
> > > They don't have to be perfect. Just good.
> > 
> > yes, it would be really great to have default settings (or automatic mode or 
> > whatever), that would give good results for all kinds of input material. I'm 
> > still wondering why Qpel is sometimes worse and often better than halfpel. I 
> > have an idea and I'm currently investigating this.
> > 
> > BTW: Could someone provide a clip where qpel performs really bad? At best a 
> > clip where for a fixed quant encoding qpel produces bigger sizes and lower 
> > PSNR. This would be helpful since most of my clips seem to be qpel friendly ;-)
> 
> Ich habe einen kurzen CLip (aus Dumbo von Disney) unter 
> 
> http://lieb.math.uni-bonn.de:8080/data/dumbo.mpg
> http://lieb.math.uni-bonn.de:8080/data/dumbo.vob
> http://lieb.math.uni-bonn.de:8080/data/dumbo.pgm.bz2
> 
> user/pass jeweils xvid
> 
> Ein typischer Cartoon mit 25fps statt 15 oder so, d.h. viele
> "doppelte" Frames, wo nur das Rauschen neu ist. Und das nicht zu knapp.  
> Auf die doppelten kommt es aber gar nicht an, im PGM-stream sind die
> doppelten entfernt, machts nicht besser.
> 
> Meine Logs (vom PGM) sind attached, dumbo1 ohne QPel, dumbo4 mit Qpel.
> PSNR ist echt auffaellig. 
> 
> Gruss,
> Christoph 
> 



More information about the XviD-devel mailing list