[XviD-devel] syskin me test in Matrix Title 29

[Christoph Lampert]

Hi Edouard,

it's good that syskin-ME has lower average quantizer. I had similar
results in some files, but that doesn't auotmatically means higher
quality, because I guess coefficient threshholding (TOOSMALL_LIMIT) 
has a different values and then average quant cannot really be compared
anymore. 

gruel 


On Fri, 20 Sep 2002, Edouard Gomez wrote:
> And now the results :
> 
> Syskin ME :
> # quant 1 : 0
> # quant 2 : 0
> # quant 3 : 304
> # quant 4 : 2483
> # quant 5 : 1588
> # quant 6 : 271
> # quant 7 : 7
> # quant 8 : 35
> 
> Average quant : 4.296
> 
> Old ME:
> # quant 1 : 0
> # quant 2 : 1
> # quant 3 : 286
> # quant 4 : 2312
> # quant 5 : 1710
> # quant 6 : 335
> # quant 7 : 11
> # quant 8 : 33
> 
> Average quant : 4.481
> 
> File size have been respected in both cases, visual quality is hard to
> judge as lot of debris are flying around. But this is a good point for
> the new  ME imo. If it's  able to reduce  quants in such a  hard scene
> that means  it performs (probably) a more  accurate motion estimation,
> or uses shorter vectors and save bits for texture data.
> 
> I forgot to benchmark speed, perhaps tomorrow with other tests. 
> 
> -- 
> Edouard Gomez
> _______________________________________________
> XviD-devel mailing list
> XviD-devel@xvid.org
> http://list.xvid.org/mailman/listinfo/xvid-devel
>