[XviD-devel] dynamic hpel/qpel decision

[skal]

	Hi Radek and all,

On Mon, 2002-12-02 at 12:10, Radek Czyz wrote:

> PS does someone feel like writing an asm-ed Sobel edge detection? It's
> not slow the way it is (copied from RefDivX's adaptive quantization
> code) but can always be faster.

	I got something like this burried in my comp-vision
	stuff. I'll try to dig it out. If I remember well,
	my filter was computing x-gradient, y-gradient and
	gradient squared norm in one shot, at reduced resolution.
	It was for edge thinning stuff 
(e.g.:ftp://ftp.inria.fr/INRIA/publication/publi-ps-gz/RR/RR-2444.ps.gz
	describes a real good method with pixel interpolation)
	Are you using edge thinning? I'd be interesting to know if
	it improves the ME matching...
	later,
		Skal