GCN on 3d point cloud data

[utkarsh0902311047]

Hi,
I would like to ask :

1. if any code is available to implement GCN on 3D point cloud data.
2. Also, which is the example tutorial for point-gnn, I cannot find it.
3. Does Point Net use any Graph neural networks?

Thank you
Regards

[rusty1s]

You can search for point* in our examples folder, e.g., here is an example of PointNet++ for classification.

[rusty1s]

PyG 2.3 is only compatible for Pytorch >= 1.11.0. I think you may need to upgrade your PyTorch version, sorry :(

[utkarsh0902311047]

So I need a higher cuda version too I believe?

[rusty1s]

Yes, that's correct.

[utkarsh0902311047]

@rusty1s
Hi
I tried running Point-GNN example by replacing PointNetConv ---> PointGNNConv.
But I am getting following error.
Traceback (most recent call last):
File "C:\Users\utkarsh\Documents\pytorch_geometric\examples\pointGNN_classification.py", line 104, in
model = Net().to(device)
File "C:\Users\utkarsh\Documents\pytorch_geometric\examples\pointGNN_classification.py", line 52, in init
self.sa1_module = SAModule(0.5, 0.2, MLP([3, 64, 64, 128]))
File "C:\Users\utkarsh\Documents\pytorch_geometric\examples\pointGNN_classification.py", line 21, in init
self.conv = PointGNNConv(nn, add_self_loops=False)
TypeError: init() missing 2 required positional arguments: 'mlp_f' and 'mlp_g'

What changes should I make?
Please help

[rusty1s]

Here is a small example on how to use PointGNNConv:

    conv = PointGNNConv(
        mlp_h=MLP([8, 16, 3]),
        mlp_f=MLP([3 + 8, 16, 8]),
        mlp_g=MLP([8, 16, 8]),
    )

so you see that it requires 2 additional MLP definitions. Hope this helps.