[Speed] Large Mesh

[asml]

I have a mesh with around 1.5 million vertices/3 million faces. Is pyoctree suitable for meshes this large? Tree creation has been running for over an hour and I was wondering if that's to be expected or if something is wrong.

[mhogg]

Hi asml,

My meshes are generally in the order of 10,000-100,000 faces, and take only a second or so to construct the octree. Your 3 million faces could slow things down, but > 1 hour sounds excessive.

One possible issue is during the conversion of your Python array to C++ array, which is handled by the Cython wrapper. If you are willing to share your code and mesh, I would be happy to have a look to find a bottle neck there or elsewhere. Another suggestion is that you decimate your mesh first to reduce to tri face count before running pyoctree.

Cheers,
Michael

[asml]

Thanks for your quick reply (and awesome project, by the way :))! You're right, after simplifying the mesh to 150k vertices/300k faces, tree construction takes just a couple of seconds. I cannot share the mesh, unfortunately, because it's confidential data, but the actual call to pyoctree is simply

ot.PyOctree(self.mesh.verts,self.mesh.faces),

where the former is a 1.5 million by 3 float array and the latter a 3 million by 3 int32 array. No surprises there. Memory shouldn't be an issue either. It seems like something doesn't scale linearly.

[mhogg]

Hi Andreas,

I will look into this in the next few days. The obvious thing here is the use of a contiguous array to hold the tree data. Currently I am appending to an stl vector, which is probably contiguous up to a certain point, but after a certain number of polys will be scattered around in the memory leading to the speed reductions you are observing.

Cheers,
Michael