RenderMan & RISpec, Smoothing normals problem

Hi (sorry for my poor english),

I've written a very simplified python script to export a basic rib from
Blender.

I use the same, unique function to convert the coords of all (objects
postions, camera position, verts, normals, etc...). All work fine except
the normals.
Or, precisely, the normals work fine with Pixie, but not with AQSIS and
3Delight...

It's strange, because if the same function convert perfectly the
geometry and the locations of objects, it should work for norms
coordinates too.

These are the results:

Pixie return the same illumination as original Blender scene (from left
side):

http://www.dedalo-3d.com/lab/pixie.jpg

AQSIS and 3Delight return a wrong illumination from opposite side:

http://www.dedalo-3d.com/lab/aqsis.jpg

If I use the file without normals (no smooth), the illumination is OK in
all engines.

I've zipped foo.rib here:

http://www.dedalo-3d.com/lab/foo.zip

It's only few lines of code...a simple test.

Anyone can check it? Pixie is the only free engine that handle correctly
the smooth normals, or the rib generated by my code is wrong and Pixie
correct it in 'realtime'? Anyone can try to render using pixar renderman?

The only error possible is a negative value instead a positive value(or
viceversa)...but, I repete, for geometry, position and world the
coordinate system conversion function work fine...
If I modify the rib using a negative value for y coord of normals
(empirically, but in theory this should be wrong), it work into AQSIS
and 3Delight, but not into Pixie (as expected).



Ciao, thanks

Manuel

The point is that normals *cannot* be transformed in the same
manner as vectors and points (using fancy terminology normals
are bivectors, not vectors!). If your transformation matrix is
M, then points ad vectors are transformed:
P |--> M.P
while normals are transformed:
N |--> (M^-1)^T . N

For details and explenation see the FAQ of CGA
( http://www.geometrictools.com/CgaFaq.pdf ), point 5.27.

Przemek

--
"Beauty is the first test: there is no permanent place
in the world for ugly mathematics." G.H. Hardy

Thanks!
However my effort is to have a human readable rib, without using matrix.
The foo.rib use only rotate and translate. For geometries and locations
the conversion is very simple. Please, take a look here:

http://www.dedalo-3d.com/lab/conversion.jpg

There are two 'conversion tables', one for translations, and one for
rotations.

Do you think it's possible using a sort of table like these, for normals
too?

Thanks,

Manuel

> However my effort is to have a human readable rib, without using matrix.

I think you can specify orientation using Orientation "lh" and
Orientation "rh" without messing with mathematics.

Felipe Esquivel

Yes...the same reply here:

http://sourceforge.net/forum/forum.php?thread_id=1296320&forum_id=200367

Now it seem to work...

Thanks!

Manuel Bastioni