[vtkusers] Problem with display accuracy for actors having large bounding box coordinates

[pof]

Hi,

I'm facing a problem with the display accuracy of actors that have large 
bounding box coordinates.
For instance :
- when I display a scene having a bounding box (0;1000;0;1000;0;100) 
everything is fine, that is all the actors look pretty nice (see 
http://www.imageshare.web.id/images/h1e5ufe0j33il4vykltr.png);
- when the same scene is just translated by a vector (1e6,1e6,0), i.e. 
its actual bounding box is (1e6;1e6+1000;1e6;1e6+1000;0;100), there are 
pretty much overlapping between actors (see 
http://www.imageshare.web.id/images/3a7eeujdvo9zntwngoz5.png).

In both case, I use vtkRenderer::ResetCamera() to fit the cam view to 
the real bbox view, but the result is not that good. I tried playing 
with SetResolveCoincidentTopology, but did not found a good solution.

An obvious workaround would be to ensure that the coord of the 
vtkpolydata behind the actors are close to the origin, but its is 
precisely what I want to avoid as the original data comes from GIS 
(Geographical Information Systems) and it is usually far from beeing 
close to the (0,0,0) point, and I would like to avoid duplicating it for 
memory efficiency.

Does anybody has an idea to overcome this problem ?
JD