Conv2d parameters as a matrix

[J_Johnson]

I’m not sure if this is what you’re looking but it seems related:

Toeplitz matrix

And this thread might help:

Obtaining Toeplitz matrix for convolution

In this issue @ezyang references an implementation of convolutions that uses the Toeplitz matrix. I have a state_dict (and also a nn.Module class) from a network and explicitly need these Toeplitz matrices for further calculations but I admittedly have not a strong grasp on the things going on in ATen and how I could use that directly in Python. Is there a way to do this?

By the way, using a Toeplitz matrix may not be the most efficient approach. This post provides a more efficient algorithm with less calculations:

Is there an function in PyTorch for converting convolutions to fully-connected networks form?

But still needs the operations to be parallelized.