[Visualpython-users] VPython 5.2

[Bruce S. <Bru...@nc...>]

Available at vpython.org: VPython 5.2 for Windows and Mac which adds a 
new feature for smoothing objects made of faces.

After creating a faces object named (say) myfaces,

    myfaces.smooth()

will average similar normals at a vertex and make the object look 
smoother. You must make sure that all vertices have nonzero normals 
before using the smooth function; a simple scheme is to make all the 
normals perpendicular to their respective faces and then let the smooth 
function adjust them for improved appearance.

The issue that this addresses is that if you try to make a 
smooth-looking curved surface but adjacent faces have different normals, 
the surface has creases in it. Smoother appearance is obtained if 
similar normals at a vertex shared by two or more triangles are averaged 
and the average normal applied to all these triangle vertices. However, 
at  a vertex shared between two surfaces with very different 
orientations, such as perpendicular surfaces, no such averaging should 
be applied. The new smoothing routine only averages shared normals if 
they don't differ too much in direction (about 18 degrees).

The example program "faces_heightfield.py" has been revised to use the 
new smooth() function.

Bruce Sherwood

P.S. Maybe someone would find the algorithm interesting, and I'll 
describe it in terms of Python code, though it is implemented in C++ for 
speed:

First pass: Construct a dictionary whose keys are vector positions of 
vertices and whose values are lists (or rather Python "sets") of indices 
of vectors in the faces pos (or normal) array.

Second pass: For each key (spatial location of a vertex) in the 
dictionary, get the value (the set of indices of pos or normal vectors 
each of which corresponds to the same position in space). Pick an index 
in this set, evaluate its (normalized) normal, and one at a time take 
the dot product of this normal with the (normalized) normals of all the 
other members of the set, creating a list of indices for which the dot 
product of the normals is > 0.95 (so the normals differ in direction by 
no more than 18 degrees). For this subset, average their normals and 
assign them to this subset, then remove the subset from the larger set 
and repeat until the set is empty.

This procedure assures that neighboring faces that don't differ much in 
orientation will not have a sharp crease between them, but neighboring 
faces that have very different orientations, including for example 
perpendicular faces, will have very different normals.

There is an inappropriate bias in this scheme in that the starting 
normal for searching through the set might be an extreme one and 
therefore one might fail to identify the right bundle of normals to 
average. Maybe someone will think of an unbiased way to do the bundling. 
I can say that having tried the smooth() function on various objects 
made of many faces, the scheme seems to work pretty well.

Another possibility would be to look at edges shared by triangles rather 
than at vertices shared by triangles. At the moment I don't see how to 
do that.