http://labelme.csail.mit.edu/index.html
It is a very interesting way to research and experiment with computer
vision.
A toolbox for a different CAS is freely available on the project site,
but apparently nothing for Mathematica.
I wondered if WRI or anybody else has contacted the LabelMe team about
developing a similar toolkit for our beloved system.
It shouldn't be too difficult to develop anyway: the current toolkit
basically downloads the database and interpret/analyse the XML code
tagging the objects present in the scene.
Considering that these functionalities are already integrated in Mathematica
and with the image processing support already present in V. 7.0x, a
powerful, yet easy to deploy, package could be created and released
free of charge.
This could be an excellent marketing vehicle to further promote Mathematica in
the image processing users segment.