Reason for using Conv3d instead of Conv2d

[MertCokelek]

Hello, first of all thank you for the amazing work and your efforts.

In the model definition in spherical_model.py, I see you are using 3D Convolutions instead of 2D, but I could not see that detail in the paper. Also, according to nn.Conv3d documentation, the kernel size order should be (depth, height, width), where you are using a kernel size of (k, k, 1) in all of the definitions.

I can not quite understand this part, is there anything I am missing? Thanks in advance.
Best,
Mert

[MertCokelek]

I dived in a little more, I guess it's because high_res_patch' has shape [bs, 3, 128, 128, 18].