Re[2]: [XviD-devel] API 3.0

[Christoph Lampert]

> On Tue, 2002-09-10 at 17:08, Radoslaw Czyz wrote:
> > BTW, do you know a way to compute PSNR of my second pass? I'd like to
> > compare hinted me search vs. full me search, but I have no idea how ;)
> > A simple program to compute PSNR between 2 avi's would be the best,
> > and I have a strange feeling that I've read about it somewhere - on a
> > forum of sort - but search at Doom9's didn't give me any effect.

E.g. with mplayer's PGM output this is just a script of a few lines in
Linux. I guess a movie player solution would be better than a new
application or a tweak to XviD, because then you are independent from
input format and/or codec. 

Skal wrote:
> 	Something close to what you want is the 'sum' returned
> 	by quant_inter() in MbTransQuantInter(). It's the SAV
> 	(yeh! a new acronym. no? ok, Sum of Absolute Values) of the
> 	inter coeffs. Would it be the SSV (ok ok. Sum of Squared
> 	Values), Parceval theorem would tell it's the PNSR of
> 	interblocks (the ones you're interested in for ME).
> 	Anyhow, this SAV is not so far from PSNR as a quality
> 	measure...
> 	This reminds me that collecting this PSNR-like stat in
> 	xvid (with a special compile flag, since it slows down 
> 	the encoding a little) is just few lines... 

There already is "internal PSNR" calculation (based on input YUV,
output YUV, not in DCT domain) as a #defin'able compile option. 

gruel