@RBD Thanks for the feedback and for the updated script. Please find below some comments:

- In order to reduce execution time, the current version of V-HACD considers only axes-aligned clipping planes. Therefore, to get optimal decomposition results, the user should apply transformations to the input mesh such that the best cutting planes are along the x, y and z directions. I’ll try improve the algorithm by extending the search to include more clipping directions.

- In the example you shared, more precise decompositions could be obtained by increasing the resolution and decreasing the maximum allowed concavity.

- Yes, open meshes are considered as surfaces and not as volumes and the ACD is computed on the surface, which may be the desired behavior for some use-cases (e.g., CAD models). If this later behavior is not desired, it is the responsibility of the user to close the holes before submitting the mesh to V-HACD.

- V-HACD (without capping the number of vertices per convex-hull) generates convex-hulls that are guaranteed to encapsulate the original mesh (i.e., bigger that the original mesh). The distance between the convex-hulls and mesh is controlled by the precision (or resolution) of the voxelization stage. For now, in order to produce more precise results, a higher resolution is needed, which comes at the cost of a higher computation time. I’ll try to fix this in future versions.

- I have added a new parameter that controls the adaptive sampling of the convex-hulls

- I have updated your latest script to include the new parameter in the UI. The latest version is available here

https://code.google.com/p/v-hacd/source ... t_vhacd.py