Polygon size

This also defines plotcolors

This defines PolySimpN, VptDepSimpN and MeshedImposterN

In a conventional mesh and assuming triangles,
each vertex is corner of 6 triangles.
In a quadmesh this would be 4 but we assume triangles throughout.
MinDistance is the minimum distance that is accepted/corrected for (in meters).
In our current implementation prototype Mesa seems unable to handle indexedfacesets with texture properly. Thus we had to 'unfold' the indexedfaceset completely, giving VertexReuse=1.

BitsPerVertex is including the normals and texture coords.
So we have 3 coordinates for the position, 3 coordinates for the normal and 2 for the texture, or 8 coordinates in total.
[Pajarola] suggests that 4 to 6 bits are required for each coordinate, giving 48 bits per vertex .
Using compressed normals using a limited set of 4096 normal directions we need only 12 bits for the normal instead of 3*6=18[], giving 42 bits per vertex.
When data is transmitted binary but not compressed, we need 4 bytes per coordinate = 32 bytes = 256 bits per vertex.
When data is transmitted ASCII, we have 10 bytes per coordinate = 80 bytes = 640 bits per vertex.

BitsPerFace is number of bits to store connectivity info. [Pajarola] comes to 3.7 bits per triangle. Without compression, 3 indices are required. Coding indices binary with 4byte integers, we get 4*3=12 bytes =   96 bits per face.Coding ASCII with 5 characters (4 digits and a space) or 40 bits per index, we have 120 bits per face.
With our prototype, the indexedfaceset mode does not work properly, giving VertexReuse=1 but no need for indices and therefore BitsPerFace=0.

We currently do not account for overhead when multiple objects are involved, transformation between objects, lighting etc.

Assuming triangle polygons..

Check with real model

Quick check, without compression. An SMF ASCII file of a cow takes 180kb for 2904 vertices and 5804 faces. 87kb is vertices, 92k is face-indices.This file has no texture and normal coordinates  estimation sould be larger than actual size. The actual vertexreuse factor in the cow is close to 6 (5804 triangles ~17412 corners / 2904 = 5.99

Check the composing parts

Connectivity estimation is exactly right (87k)

Vertex size is too large:

Omitting per-vertex normal and uv coord, as is the case in our test file, saves 5 of the 8 values, so multiply by 3/8. Then the approximation is quite good (92k in our file):