Cannot pass weights of R2Conv as a positional argument

[quva-lab]

Cannot pass weights of R2Conv as a positional argument

[Gabri95]

Hi @Branikas

As far as I know, PyTorch's torch.nn.Conv2d also only accepts the input tensor in its forward() method (see here).

Is it possible that what you mean is torch.nn.functional.conv2d?
That, however, is not a torch.nn.Module, but a simple function.
We do not have an equivalent function inside our library since R2Conv needs to first expand the weights to construct the convolutional filter. However, this operation requires some additional information, which is generated inside the R2Conv.__init__ and which is stored inside R2Conv.basisexpansion.

Hope this helps

Gabriele

[Branikas]

Hi Gabrielle,

Many thanks for the quick response. Yes, it actually makes sense. Thanks a lot for this.
Feel free to close the issue now.

Regards,
Stathis