[XviD-devel] GMC

Christoph Lampert xvid-devel@xvid.org
Tue, 31 Dec 2002 00:00:40 +0100 (CET)


On Tue, 31 Dec 2002, suxen_drol wrote:

> 
> On Mon, 30 Dec 2002 20:49:15 +0100 (CET) Christoph Lampert <chl@math.uni-bonn.de> wrote:
> 
> > Considering that maybe not too much changed in GMC encoding,
> > what's the status of decoding? Anyone? 
> 
> hi christoph,
> 
> the decoder supports gmc (1 warp point; translation).
> what do you mean by "4 parameter" model... are you using 4 warp points?!

No, but 2 warp points (= 4 free parameters in total).

I don't like 1 warp point because there's no real difference to normal
motion vectors, and you can gain hardly anything from GMC if it just
implements stuff already implemented just in a different way.
That's one reason why current GMC has no positive effect. The other one
is, that's the implementation is bad, of course... 

But with 2 warp points you can encode zooms and rotation, that might save
a few bits (e.g. during zooms and camera rotations ;-)

3 warp points might not be necessary, because the extra features
(coding shears) are rarely used in real life video. 

ffmpeg can decode up to 3 warp points, so I can test encoding even without
XVID support for decoding... but I wanted to make sure what the status is. 

gruel