[XviD-devel] gme code

[Christoph Lampert]

On Thu, 13 Nov 2003, elcabesa wrote:

> On Thursday 13 November 2003 19:23, Christoph Lampert wrote:
> 
> > P.S. The wrong code should only very slightly change behaviour, by
> > choosing less frame for GMC. If you had problems before, you will still
> > have (more) afterwards.
> 
> i have problem iwth GME not gmc..
>  i still couldn't find a error in code but i don't know too much algorithm you 
> use, i only could search for visible wrong code

Actually, it's very simple. Gradient decent in 6-dim. GMC parameter space.

Start with some parameters. Calculate GlobalSAD. Change all parameter by
+1 or -1. Everytime calculate GlobalSAD. Choose the best of all 7
possibilites (no change, +-1 in 3 components). Update current best. 
Repeat, until the position doesn't change anymore.

gruel