[XviD-devel] Fixed quant second-pass
chl at math.uni-bonn.de
Mon Dec 15 11:39:02 CET 2003
On Mon, 15 Dec 2003, Radek Czyz wrote:
> Hi :)
> Christoph Lampert wrote:
> > I still have the impression that the "clever" twopass varies quantizer too
> > much. +-1 of average could absolutely be enough.
> I think DivX does it this way - at least their n-pass appears to keep
> constant average quantizer, more accurately with each pass. Imho it's a
> good solution - but XviD's might better, because if frame takes very
> little bits or at least its textures take little bits (with newest Gom's
> experiment), it's given quant 2, as it doesn't make any difference anyway.
If it takes few bits, changing to quant 2 doesn't change image quality
much, either. I would just like to have a "rock solid" 2- or N-pass, where
even the strangest framesize distribution cannot cause any _really bad_
frames, even if that means that average PSNR drops by 0.02 or so.
More information about the XviD-devel