How to combine surfaces from chunks (or how to deal with a real large region)

[HHHit]

I am trying to generate the surface mesh of a very larger object (the bounding box is around 8000x8000x1400 voxels), which requires a very large memory. I am wondering if there is a way to combine the meshes generated from each chunck? (like 800x800x140 voxels) Thanks

[william-silversmith]

Hi HHHit! I work on another system that is exactly for this problem. Check out https://github.com/seung-lab/igneous if you are working in the neuroglancer ecosystem. Otherwise, you can use cloudvolume.Mesh.concatenate

[HHHit]

Thanks for replying! Currently, I am trying to generate a .obj file for the object. Just to make sure, I can first use zmesh to generate the surface mesh for each chunk (from binary 3D segmentation) and then concatenate them using cloudvolume.Mesh.concatenate, is that right? I quickly looked into the code of cloudvolume.Mesh.concatenate, it seems to simply combine the array of vertices, faces. But what about that faces between chunks, can it also connect the vertices between chunks?

[william-silversmith]

If you make sure to mesh with 1 voxel overlap, the edges will line up exactly. You can then use the cloudvolume calls to remove duplicate vertices which should yield a proper mesh.

[HHHit]

I am still confused about how to transfer from Zmesh to cloudvolume.Mesh. What if I obtained the faces and vertices of the meshes from two chunks separately, how can I use cloudvolume.Mesh.concatenate to combine these two?

[william-silversmith]

The zmesh Mesh is code that's copied from cloudvolume and stripped down. Just do:

from cloudvolume import Mesh
m1 = zmesh.get_mesh(...)
m1 = Mesh(m1.vertices, m1.faces)
...
mesh = Mesh.concatenate(m1, m2)
mesh = mesh.consolidate() # if you don't have self-contacts
mesh = mesh.deduplicate_chunk_boundaries(chunk_size=(512,512,512)) # if 512,512,512 is your chunk size