Re: [openvrml-develop] Re: Transparency

[Christopher K. S. J. <ck...@di...>]

Damian McGuckin wrote:
> 
> > Sort them once at the begining.  Then resort on every single change in
> > viewpoint. Assume that as you move around the list will remain mostly
> > sorted, and choose a sorting algorithm to fit that assumption.
> 
> If you rotate yourself, say 90 degrees, doesn't the list get horribly out
> of sort.  Mind you, this is simply a Z-buffer problem.
> 

 The theory is that there should be many intermediate
steps in between, each of which results in only a small
numbers of postion swaps in the sorted list. Spatial
coherency is a fairly common assumption for real time
scene graphs, even though, as you point out, it's not
always true.

 Of course, it's only the transparent surfaces that
need to be depth sorted, so as long as you keep to
a reasonable number of them, the sorting shouldn't be
a big deal.

 A good intermediate step might be to keep all the
transparent surfaces in an unsorted list and render
them last. That's not quite right (transparent
surfaces layered over other transparent surfaces
would have subtle artifacts), but would be much better
than the current situation (transparent surfaces
layered in with any surface at all produce very ugly
artifacts)

 In the bad old days when processor time was expensive
and 3d harware was rare, you also had "screen door"
transparency. It's ugly, but relatively easy to 
implement, since you don't have to keep a separate
(sorted) list of the transparent surfaces. A google
search should turn up a bunch of references.



-- 
Christopher St. John ck...@di...
DistribuTopia http://www.distributopia.com