|
 |
I thought I'd bring the discussion about patch vs mesh over here. I got
interesting results doing some tests of the two using the same model.
type 0 patches are light on memory but take extremely long to render.
type 1 patches are much faster, but the memory goes through the roof.
A mesh of plain traingles suprised me, as the render time were longer
than smooth traingles. Parsing was shorter as expected and the memory
use was a little lower. Triangles in general seemed to be faster on the
trace than patches, but the parsing time gets really long. Nathan also
pointed out that the mesh I was using was 1/4 as dense as it should have
been.
The type 2 patches in the superpatch seems to have the best balance of
parse and trace time, and since they don't convert the patches to
triangles prior to tracing, the memory used is quite low.
One thing I was hoping someone might have some info on is uv mapping for
bicubic patches. I have a working exporter for animation master and
there are uv coordinates available in the export, but currently I don't
have anything I can use it with. The way it works in the uv mapping
patch makes sense, but other programs I've looked at specify 4 corner
vectors for uv mapping. I assume this means you can have a section out
of the image that isn't rectangular. I want to try to add this to
POV-Ray if I can.
To make matters worse, I realized that animation master uses 12 uv
vectors to specify the mapping. These are all the outside points,
excluding the 4 inside the patch. The RIB format only supports 4
vectors, and this is really obvious when I export to that format.
I assume that the way A:M does this is to somehow project bezier curves
onto the image and extract the values. Anyone ever hear of something
like that before?
If I can get this to work it would be so cool combined with the type 2
patch method in the superpatch, since it should render very fast. I
should be able to add animation capabilities eventually as well.
-Mike
Post a reply to this message
|
|
