[XviD-devel] quick question about the quality switch

[MAILER-DAEMON]

I had a quick question about this.

Wouldn't quantization influence more on quality then ME?

Better ME might give you a better estimation but I would think
quantization would effect the quality of the decoded image.

I agree that for same quantization factor that a better ME should
generally yield smaller encoded size of the frame but I would think the 
quality of the decoded frame would not effected as much as the 
quantization.

I'd appreciate any feedback.

On Thu, 5 May 2005, Christoph Lampert wrote:

> 
> Hi Tom,
> 
> it might be too late for your paper, but this is the explanation of the 
> quality switch: 
> 
> The setting indeed mainly relates to the motion estimation (ME). 
> But, as you will find out when doing a little more research on 
> XviD, we of course don't use use full search to find the MVs, because 
> that would be much too slow. We use a selfmade and -tuned version 
> of the "zonal algorithm" idea by Alexis Tourapis, which is kind of 
> a mixture between PMVfast and EPZS. There is no "search window" that 
> could be influenced. But there are many other parameters than can be 
> switched on or off, which you will find in the XviD source at src/xvid.h 
> and e.g. in vfw/codec.c you can check which are used when. 
> 
> The higher the quality level, the more of these features are switched
> on, e.g. 
> * refining to halfpel position instead of just fullpel 
> * using 4 instead of just 1 motion vector per 16x16 macroblock
> * using square patterns instead of diamond pattern for the ME
> * use several starting positions for gradient descent instead of 
>   just one 
> 
> To explain how these work in detail would not be a "brief description" 
> anymore. 
> 
> gruel
> 
> 
> 
> On Wed, 4 May 2005, Tom Jacobs wrote:
> > could someone give a a very brief desciption about the differences between
> > -q 1 to 7? i need to add it to a paper and its in today :$ i have (well
> > someone else in my lab who has left) profiled the encoder at different
> > levels and now the reviewer of the paper wants a line or two saying the
> > difference.   im guessing the levels relate to how good ME is?  xvid only
> > uses full search to find MV though doesnt it? so does this just effect the
> > search range used? are there any other differences?
> > 
> > many thanks
> > 
> > Tom
> > 
> > _______________________________________________
> > XviD-devel mailing list
> > XviD-devel at xvid.org
> > http://list.xvid.org/mailman/listinfo/xvid-devel
> > 
> > 
> _______________________________________________
> XviD-devel mailing list
> XviD-devel at xvid.org
> http://list.xvid.org/mailman/listinfo/xvid-devel
> 

--------------------------------------------------------------------------
   Sang-Uok Kum                    Dept. of Computer Science
   kumsu at cs.unc.edu                Sitterson Hall CB #3175
   http://www.cs.unc.edu/~kumsu    Univ. of North Carolina - Chapel Hill
   office: (919)962-1869           Chapel Hill, NC  27599-3175
--------------------------------------------------------------------------