Filters should be done via convolutional theorem at larger scales

[leios]

If the filter sizes get too large (~100x100 or larger), it makes more sense to use the convolutional theorem:

ifft(fft(A).*fft(B))