octree-nn / ocnn-pytorch

Octree-based 3D Convolutional Neural Networks
MIT License
156 stars 17 forks source link

Convolutional layer dimension #4

Closed samanemami closed 2 years ago

samanemami commented 2 years ago

As I am seeing, this library only uses the 1D Conv layer in the module. Is there any other implementation with higher dimensions for the layers?

wang-ps commented 2 years ago

Although the feature is stored in a 1D tensor, the Conv layer is actually 3D. ocnn is a 3D sparse convolution framework.

samanemami commented 2 years ago

Thank you But could you please explain it in the code?

I can not understand how would it be a 3D layer as the input has not had four dimensions!

For instance, in this line what is the definition of the convolutional layer? I do not know the dim!

I am not sure about the 3D layer; all the layers I am looking at them are 1d same as the following;

self.bn = torch.nn.BatchNorm1d(out_channels, bn_eps, bn_momentum)
wang-ps commented 2 years ago

Please read our paper and Slides to understand the Octree-based sparse convolution.