[XviD-devel] Request for ideas/comments about 2pass

[Christoph Lampert]

On Sun, 25 May 2003, Edouard Gomez wrote:

> Hello,
> 
> I've been  digging deep into the current  2pass code and had  run lot of
> tests now.
> 
> On the one hand, i can say  that the file size control works. It is very
> accurate ... I had never undersize/oversize surprises. 
> 
> On the other hand, the current implementation is far too aggressive with
> bframes. I remember we talked a lot about the fact that a linear formula
> to compute a bframe quant from a  pframe quant was not a good idea. That
> was at that time that we decided to introduce the offset/ratio parameter
> couple.
> 
> But this file (http://ed.gomez.free.fr/vrac/pass2.txt) shows clearly the
> current 2pass code is acting exactly like we wanted not. It uses far too
> high quants for bframes compared to their reference frames.

Twopass sets the quants of bframes directly and doesn't calculated it from
surrounding P-frames??? That can't be good... I'm sure B-frames scale
differently than Pframes, so they should either get their own routine or
their values just calculated by offset/ratio. 

gruel