Healing/rebuilding multiple meshes in .stp file

[VoRodin]

Hello, first of all, thanks a lot for developing this library and MeshInspector.

Now I am facing a problem where I have 2 meshes, where one should be healed/rebuild
absorber_w_plate_2mm.zip

I have tried the heal function from healer.py

from meshlib import mrmeshpy as mm

mesh = mm.loadMesh("absorber_w_plate_2mm.stp")

new_mesh=heal(mesh,0.1)

But it took a long time and didn't finish.

Another question mark is how to find the optimal voxel size for rebuilding it using MeshLib.

Thanks

[Grantim]

Hello!

We have function suggestVoxelSize to find voxel size to approximately fit desired number of voxels. Basically voxel size represent precision of rebuild.
In MeshInspector we use
mm.suggestVoxelSize(mesh,1e9) as minimum voxel size
mm.suggestVoxelSize(mesh,1e4) as maximum voxel size

mm.suggestVoxelSize(mesh,1e8) as default voxel size

but also please note that in MeshInspector we use CUDA implementation of the algorithm (if possible) that gives major speedup compared to CPU version. (We will provide CUDA algorithm in python in future, but now it is not available).

from meshlib import mrmeshpy as mm

mesh = mm.loadMesh("absorber_w_plate_2mm.stp")

new_mesh=heal(mesh,mm.suggestVoxelSize(mesh,1e7))

---

P.S.
I could not reproduce your issue, on my PC your code worked ~ 10-20 sec finished

[Grantim]

Let me clarify:
MeshInspector with CUDA enabled should work faster
MeshInspector with CUDA disabled should work with the same time (or a bit slower due to graphics) as meshlib python module

[VoRodin]

Ok, then there is something wrong. With CUDA I don't see much speed-up. It was installed from the Windows installer package
I also forced it to run on the discrete GPU

[Grantim]

Thanks for reporting, we will investigate it

[Grantim]

Just found a reason, really it is all valid:
As far as your mesh does not have holes CUDA does not perform much difference. So all three (MI CUDA, MI CPU and ML) work about the same time (15-17 sec on my PC with voxel size 0.1 and decimation enabled, 5-7 sec without decimation).

P.S. CUDA gives more advantage in healing meshes with holes

[VoRodin]

Built MeshLib from source and heal function worked very fast!👍