Is there a built-in function in any torch-compatible library that allows to consider convolutions (in 2 dimensions) with kernels that depend on the coordinates of the input.
