problems with visibility computation on MIT floorplans

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

in my experience, vis problems have often been due to

   1) missing or non-axial macro polygons
   2) gaps or interpenetrations in axial macropolygons
   3) suboptimal choice of the "grid spacing" parameter
   4) axial subdivision gone awry (too few/too many splits,
      or splits occurring in bad places e.g. in free-space)

some suggestions:

   make sure all the layers containing axial structural
   polygons (walls, doorjambs, floors, ceilings) are
   including in the macro polygon list fed to the axial
   subdivision stage and the visibility computation stage

   make sure you're processing the floorplan at a stage
   during which all its macro polygons are axis-aligned
   (in our system, before rotation to campus coordinates)

   check grid parameter (this controls the coercion of
   polygon vertices to multiples of the grid spacing).  for
   the berkeley models, i believe units were inches, and the
   grid spacing was set to 0.25 (so coercion to 1/4 inch).
   but for MIT data aren't the units feet?  if so then the
   grid spacing should probably be 1/12 or 1/48 or the like.
   whatever you set it to, ideally it should be exactly
   representable as a single-precision floating point
   number, i.e. it should be a small integer multiple of
   a (possibly negative) power of two.

   check the subdivision on a few test examples by suppressing
   the vis computation stage, and simply flying through the
   whole model with poly, cell, and portal display turned on.
   are there cuts along all large (area) polys?  (there should
   be.)  are there many cuts in free-space or through small
   (area) polygons?  (there shouldn't be.)  are the "big"
   horizontal cuts (the floor separators) happening before
   floor-specific vertical cuts?  (they should be.)

Michael Craig wrote:
> there were a few problems with visibility but we got at least one MIT
> floor working a few nights ago. i'll try to get everything going through
> it and make sure it works tomorrow
> 
> michael
> 
> On Wed, 23 Apr 2003, Seth Teller wrote:
> 
> 
>>if possible, i'd like to see some progress on the
>>effort of getting MIT-extruded models running
>>inside the walkthrough code tomorrow.
>>
> 
> 
>