[vtkusers] tensor glyph, inward pointing surface normals

[.]

I think I've answered my own question:

no. the inside out glyph surface with inward pointing surface normals 
does not provide useful information about the tensor field, so the 
phenomenon is a bug.

my reasoning is this:

In the vtkTensorGlyph algorithm the eigenvectors become the columns of 
the rotation matrix. Eigenvectors are unique up to a constant, so if you 
took any eigenvector and multiplied it by -1 it's still an eigenvector. 
Doing so to only one of the eigenvectors would in the simplest case 
(positive eignevalues and no further translation or rotations applied) 
change the sign of the determinant of the transform, resulting in 
inverted surface & normals, having nothing to do with the properties of 
the tensor field, and only because of the choice of the eigenvector. So 
an inside out glyph surface with inward pointing surface normals can't 
generally be used for insight about the tensor field.

On 05/11/2011 01:04 PM, burlen wrote:
> Hi all,
>
> I recently started to experiment with vtkTensorGlyph and found that 
> some of the glyphs apear as "black bodies"(see attached images). This 
> happens because their surface normals point inward. I noticed that if 
> the transform applied to the glyph has a negative determinant the 
> glyph surface and its surface normals get turned inside out, and that 
> I could "fix" these flipped surface normals by applying a scale of 
> -1,-1,-1 to the transform.
>
> I'm not sure if these inward pointing surface normals are really a 
> bug. It sure looked like one at first, but the fact that the glyph 
> surface is turned inside out is, maybe, useful information about the 
> tensor field. Although, If one is using a black background the these 
> glyphs disappear, which is not useful, and without shading they can be 
> tricky to "see" the shape at different angles.
>
> Is this a bug or a feature? Is the inversion of the glyph surface 
> useful information?
>
> I filed a bug report about this, with a more images, patch with the 
> flipped normals "fix" and two other fixes (one for parallel use, and 
> another for naming output arrays), and a small test script that 
> demonstrates the behavior. 
> http://public.kitware.com/Bug/view.php?id=12179
>
> Thanks
> Burlen
>