How to work with 3D Mesh files and data?

[royvelich]

Hi,
I am developing a GNN framework to process 3D mesh files.
So, I have a collection of OBJ files, which I load with the helper function read_bj which is provided by the torch_geometric.io module. The data object that is returned by read_obj has two main attributes: pos and face, while the edge_index attribute is missing. Now, what I don't understand, is how to use that data object in my convolution layers. Note that the MessagePassing.propagate requires an edge_index argument, which in m case, is missing.
I would love to hear how do you handle 3D meshes in your scenarios. Is there any example I can follow?

Thanks.

[rusty1s]

You can use the FaceToEdge transform to convert your face representation to a graph.

[royvelich]

Thanks.
I also see that get_mesh_laplacian returns an edge_index. Is it equivalent to what FaceToEdge returns?

[rusty1s]

Yes, that is indeed similar. It will also give you an edge_weight corresponding to the Laplacian.