The image files in this directory are in KBVision format and
are all of size 256 x 256. If you don't use KBVision, then
strip off the header. Each grey-level image, fname.im, is an
8 bit image, while the image, fnameTH.im is the thinned contour
image resulting from a Canny operator applied to the image followed
by skeletonization (to ensure 1 pixel wide contours) and gap closing
(Beymer algorithm). The thinned image is 1 bit per pixel.

The images contain scenes containing geon-based objects. The two
images, lamp5.im and lock2.im, are real images of real objects, while
the rest are synthetic images generated from Sandy Pentland's
Thingworld modeling tool.

If you have any problems, please contact me at sven@vis.toronto.edu

to acquire the images, type

ftp ftp.cs.toronto.edu

login to "anonymous"

and cd to vis/sven/images

We would appreciate it if you ftp images during off-peak hours
(our time).

Thanks,

Sven Dickinson