Exposed simplify_mesh_lossless method

[Kramer84]

This PR exposes the simplify_mesh_lossless method, which appears to be faster and more effective compared to the standard method with the lossless flag enabled. It also behaves differently in terms of triangle reduction and convergence. Additionally, the threshold_lossless argument in the standard method does not seem to influence the results as expected—triangle counts converge to a fixed value regardless of the threshold. However, in the simplify_mesh_lossless method, the epsilon parameter has a clear effect on the result.

This new method may be especially useful in batch processing scenarios where multiple meshes with similar characteristics need to be simplified efficiently.

On a minor note, the example comments at the end of the Simplify.pyx file have been removed as present in the readme. And it bugs me as to why this hasn't been done before, as I may have a mistake in closing this previous PR ...

Observations:

1. Standard Method with lossless Flag:
   The lossless flag, when used with different threshold_lossless values, results in minimal triangle reduction regardless of the threshold value.

   • Example tests:

     simp.setMesh(bunny.vertices, bunny.faces)
     simp.simplify_mesh(preserve_border=False, lossless=True, threshold_lossless=1e-3)

     Output:

     iteration 0 - triangles 112402 threshold 0.001
     ...
     iteration 95 - triangles 112402 threshold 0.001
     simplified mesh in 0.9819 seconds from 112402 to 112400 triangles
     

     Similar results were obtained for higher threshold values (e.g., threshold_lossless=10), with the mesh consistently reducing to 112380-112400 triangles.

2. Basic Exposed simplify_mesh_lossless Function:
   This method simplifies the mesh more significantly and efficiently compared to the lossless flag approach.

   • Example test:

     simp.setMesh(bunny.vertices, bunny.faces)
     simp.simplify_mesh_lossless()

     Output:

     lossless iteration 0
     ...
     lossless iteration 5
     simplified mesh in 0.1606 seconds from 112402 to 59960 triangles
     

3. simplify_mesh_lossless with Different epsilon Parameters:
   Changing the epsilon parameter in simplify_mesh_lossless clearly influences the degree of simplification:

   • Epsilon = 1e-9:

     simp.setMesh(bunny.vertices, bunny.faces)
     simp.simplify_mesh_lossless(epsilon=1e-9)

     Output:

     simplified mesh in 0.0851 seconds from 112402 to 112124 triangles
     

   • Epsilon = 1e-3:

     simp.setMesh(bunny.vertices, bunny.faces)
     simp.simplify_mesh_lossless(epsilon=1e-3)

     Output:

     simplified mesh in 0.1277 seconds from 112402 to 59960 triangles
     

   • Epsilon = 5*1e-1:

     simp.setMesh(bunny.vertices, bunny.faces)
     simp.simplify_mesh_lossless(epsilon=5*1e-1)

     Output:

     simplified mesh in 0.1603 seconds from 112402 to 9466 triangles
     

Conclusion:

The simplify_mesh_lossless method offers a faster and more predictable alternative to the standard simplification with the lossless flag. For cases involving large batches of meshes or cases where consistent simplification is needed, it may be a more suitable option. Further testing could explore whether this method could replace the current lossless flag implementation in certain scenarios, without creating an other method.

PS : Sorry for the trailing white spaces, sublime text removes them automatically. >< Only the new cdef and class method have been added.

[bwoodsend]

Looks good to me. We'll want to adjust the verbose handling to match the changes from #28.

[Kramer84]

Normally I've adjusted the verbose handling for the other method. Maybe before merging the readme file should be updated with the logging + the new lossless method, and maybe a link to this PR for the testing data.

[bwoodsend]

Yeah, I think it's important that the logging thing gets mentioned or, given that the debug messages are suppressed by default, nobody is even going to know they're there.

And yes, an example of simplify_mesh_lossless() too. I'd say no though to a link to the PR – people shouldn't need to hop around GitHub to learn an API that's essentially just two functions.