Re[2]: [XviD-devel] Other clip with Bframes, quality vs size

[Christoph Lampert]

On Mon, 19 Aug 2002, Radoslaw 'sysKin' Czyz wrote:
> > bquant_ratio (Dogma-bquantratio.png). Here the colors mean:
> 
> > green: bquant_ratio = 100    result: always worse then non-bframes
> > blue:  bquant_ratio = 150    result: always better than non-bframes
> > purple:bquant_ratio = 200    result: even better in some cases
> > turquoise:    ratio = 250    result: best, in many cases (!)
> 
> I guess you used my code. So, I have a question: did you alter
> SKIP-threshold for B-frames? 

Yes, but don't ask me to which value. I'll look it up when I come home.

> I have no idea how big this threshold
> should be. I set it to 100 at the beginning and didn't see any
> difference. To be safe, I decreased it to 70 - similar threshold for
> P-frames is only 10 (well not really, it's for initial decision only).
> 
> possibly good results for high quant simply mean that this threshold
> should be higher. I myself encoded a scene with Threshold = 180 and it
> looked still the same to me (and the sizes of B-frames were below 500
> bytes at quant 4 & resolution 640x400).
> 
> Do you have a working PSNR-computing thingy for B-frames? Do you think
> you could send it to me?

Yes, when I come home. 

> I tested b-frames with anime a bit. The (subjective) results are:
> When there is a scene with only camera panning (common in animation) -
> filesize is 25% smaller *and* 'visiual quality' is higher - there is
> no visible static noise on smooth background, while it's quite visible
> with p-frames only. I definitely like it.
> For other scenes it was good, too. But not that good.

> > So, what parameters to choose for an arbitrary file?
> > I don't have the faintest idea...
> 
> Dynamic P/B decision sounds great. However, I have absolutely no idea
> how would that work. Do you have some ideas? Or papers about it?

No, that's my own idea. I got the idea because 
a) sometimes Bframes are good, sometimes not
b) the already _have_ a routine to encode Bframes as P, namely in DX50
mode. So it wouldn't be too much work.

gruel

-- 
Christoph H. Lampert chl@math,uni-bonn,de | Diese Signature wurde maschi-     
Beringstr. 6, Raum 14 Tel. (0228) 73-2948 | nell erstellt und bedarf
Sprechstunden: keine, aber meistens da    | keiner Unterschrift. AZ 27B-6