[MPlayer-dev-eng] samples which look bad when encoded with lavc 2-pass needed

[D Richard Felker III]

On Mon, Jul 22, 2002 at 01:48:28AM +0200, Michael Niedermayer wrote:
> Hi
> 
> On Monday 22 July 2002 00:07, Arpi wrote:
> [...]
> > Hmm. I don't think that divx4's (xvid use the same) 2pass code is better...
> > i've hacked it too much to know it's shit. Imho your code is far better,
> > _but_:
> > - no one experienced with good values - defaults aren't the best imho
> i know, the defaults are mostly just randomly choosen without much testing, 
> ill fix these (vqmin should be the lowest bugfree quantizer depending upon 
> the used codec, ...)

How about fixing msmpeg4 so qscale=2 works? If it's not possible to
override qscale on a per-mb level when there are overflows, how about
just re-encoding the whole frame at qscale=3 if when 2 generates an
overflow? That should fix the issue and make higher quality encoding
possible too.

> > anyway i also experienced - and read on -users too - a possible bug:
> > when the input media is too low quality, so there are parts where the
> > maximum reachable bitrate (due to high qmin or missing data) is under the
> > wanted one (calculated by 2pass algo), it doesn't (or at least it seems it
> > doesn't) distribute these 'free' bits for the remaining part of the movie.
> someone should upload such a file

I think any movie with lots of very low motion scenes and some
high-motion ones should demonstrate the problem, if you tune the
bitrate appropriately. Ghost in the Shell DVD is my favorite example
of the problem.

Rich