next steps

[Seth Teller <seth@graphics.lcs.mit.edu>]

andreas,

now that BMG is working, here are some recommended 
next steps for you.  

first, try to construct an interactive "session" 
with BMG that exercises each of its capabilities.

for example, starting with a 2D floorplan with some
errors (perhaps rick lewis, cc'ed, can point you to
a good candidate), you can import it into BMG, then
for each of BMG's menu picks/actions, indicate an
example structure in the floorplan and use the BMG
action to correct it.

second, you should also explore the extent to which 
BMG, in its current form, can find and fix "most" of
the errors in the original floorplan, without explicitly
being told to do so by the user.

in this way you can generate a *concrete* example 
of how to use each of BMG's geometric routines, both
batch and GUI-based.

our ultimate goal, of course, is to suppress the user
interface and apply BMG's subroutines directly to a
vertically-aligned set of 2D floorplans, in batch mode.
keep this in mind as you explore the tool's capabilities,
read and re-read rick's thesis, and of course as you
talk to rick directly.

note that i wrote "suppress" the GUI, rather than 
"remove" it.  in my view, we should preserve and extend
the current GUI function of BMG for development and 
testing purposes -- interactively applying new correction
techniques is the ideal way to see if they make sense,
and if they actually work, on small examples.  once 
they have been proven to work, we can simply divorce the
correction routine from the GUI, and invoke the correction
routine in batch mode.

so third, you can start to formulate the task as a series
of predicates which locate 2D elements in the input, and
a series of operators which correct, extrude to 3D, etc.
those elements into their corresponding 3D elements in 
the output.  in the case of BMG as it now exists, those
predicates are partially algorithmic, and partially 
provided by the operator through the GUI.  in our case,
they will have to be fully algorithmic, and whatever we
can't detect algorithmically, we will report to the 
authors of the floorplans so that they can supply 
further, disambiguating information in the source floor-
plans.

prof. t.

ps i'm cc'ing prof. sequin also, so that he and rick
can add their words of wisdom on this subject.  finally,
i'm cc'ing the BMG group so that this message will be
archived at http://gfx.lcs.mit.edu/bmg/threads.html .
any messages which cc "bmg@graphics" will be similarly
archived.