estimado Loic,

i don't have an idea of what you men by 'Jean Yves implementation'.
i never saw it posted anywhere
and where is the code?

apart from this we try to make things easy for the people
and not a head-breaking library like gridflow is.
and that can only be used by scientifics..

tom schouten already went  in that direction
and i totally disagree with it

ciao,
sevy

Loic Kessous wrote:

Dear Lluis and Yves,

I'm reading your paper about openCV from the pdconf, and I was thinking about a few things that I'd like to share with you.

Some month ago I work on a project with pd and openCV and I observe that there was quit different approaches...

with different advantage regarding platforms and implementations, including the use of flext o not apparently also.

I recompiled and test for different purposes those different ones: pix_ , pdp_ and the one from Jean-Yves Gratius, and also some new experiment from myself (which the idea was to use openCV for display and video acquisition, this works more or less depending on the platform and configuration).

...this on MSW, Ubuntu and os X..., which tend to show that the cross-platform issue is not so obvious :)

one very interesting thing with Jean-Yves approach is that from one external to another only a pointer (the one of the OpenCV image) is passed, and not the full image as (except if I'm confusing) with pix_ or pdp_ . But this need  to compile the external as a full library (well it is like this in his implementation, I don't know if there is a issue to avoid this), and understanding the code and writing new codes is diiferent, I would say more complicated.

This may make things more complicated for people who want to improve the library but just want to add new externals and not deal with all the lib.

He use code:block to compile it, which is both a advantage and an inconvenient:

it took me more sometime to understand it, make link to flex, pd, gem...but then it is in a way comfortable, and it is cross-platform.

I wonder if it should not be possible to keep the idea to pass only the openCV image pointer (much more efficient) but using an approach closer to pix_ or pdp_ . From another project, I heard that there was some work about using shared memory in max and pd for similar issue. do you have any idea about this?

Best,

Loïc