implementation question

[m-lyon]

My apologies as this isn't an issue per se.

I stumbled upon this implementation after attending a talk about the Group Equivariant CNNs, and have read a bit of the paper behind this. I'm a little confused about how the striding/sliding of the filter is implemented here.

The diagram you provided suggests (though i think i've misinterpreted this), that in the 8 degrees of rotation case you're splitting the input image, say 20x20 into 8 triangles (made up of 4 quadrants which in this case would be 5x5, each containing two triangles), and you convolve a rotated filter with each of those subregions? Is that correct? If that is the case then is your kernel triangular too? Is there any sliding of the kernel across the image going on?

My apologies but i'm quite confused as to the basics of what's going on here