[SIP-devel] hello & some thoughts

[Ricardo F. <ric...@us...>]

Hi all,

It has been a while, but I've always kept SIP in mind.  I'd like to
share with you some ideas motivating me towards this project. They
should also be useful for SIP forks and other similar software:

Developing SIP is not only towards generating a final useful toolbox;
this development activity can fundamentally be seen as a *fun* way of
playing with image processing and computer vision algorithms. I can
argue that this Fun is so deep and far-reaching that it has power to
revolutionize computer vision as a field - and to make it
understandable, to make it more applicable, to popularize it, to teach
it, and to develop new methods for it. Ok, that is all beautiful and
good and abstract, and so here are some pragmatic ideas from this
"Just for Fun"/Torvaldian way of thinking:

It would be cool to have SIP be explicitly divided into two:
     1) a core SIP which is built by default and should be as stable
as possible. This has the universally useful image I/O functionality
(imread, imwrite, imshow, imfinfo, ...), and perhaps other
easy-to-support stuff.
     2) an experimental SIP, which is all the rest of functionality.
This should only be enabled by passing an explicit configure flag e.g.
:
                 ./configure --all_fancy_features


The experimental component would take the Just for Fun philosophy to
an extreme. More specifically, it would ruthlessly interface to as
many external libs as desired (VXL, OpenCV,  Leptonica, LTILib,
AnImaL, and so on), without much concern about portability or ease of
installation at first.

As a developer, it is fun  to write Scilab/scripted interfaces to
these external libs as it helps you learn the algorithms and
functionalities of these libs, and also helps you have this
functionality easily available in a nice easy-to-use environment. You
then see your work being used by those power users like adventure and
use  the "--fancy_features" functionality, and with time these can
even make into 'core'. Note that even if Scilab itself has its own
"quirks", that itself actually adds to the fun and can actually lead
to different and innovative ways of exploring image processing.

Writing Scilab interfaces to these libs or even scilab scripts for the
SIP toolbox is also an awesome way not only to explore these
algorithms, but also to document them. Each function you write for SIP
is documented with a nice interactive example, and can also be part of
a demo. That's the other thing, producing demos in Scilab are
extremely easy and fun to do once the scripting interface has been
built. And having demos just makes computer vision more fun.

So out of these purely "just for fun" activities, I believe other good
things come as a natural consequence. Naturally, a community starts
being built around SIP due to the social culture of free software
based on sharing. The computer vision field moves forward. Scilab
itself gets better.
Image processing courses get more fun to teach and participate. It all
works well, as long as we don't start "pretending" SIP is a finished
product to be pristinely packaged & delivered & supported, but just
accepting that above this it is just another way of having fun with
algorithms, computers, and vision.

PS: This is not just talk.. I hope. I'm installing Scilab + SIP on
Ubuntu Lucid & Mac OS X right now. Join the fun!
Ricardo