STEGO
STEGO copied to clipboard
Confused about the spatial centering implementation
As per my understanding, in this code for spatial centering, the operation fd -= fd.mean([3, 4], keepdim = True) should have been enough.
Can you please correct my understanding/offer intuition into this segment of code - especially why fd = fd - fd.mean() + old_mean is done? Thank you.
I have the same question !
It's been a little while but IIRC its so that you dont change the aggregate strength of the signal, aka the full mean (here old_mean). We still want positives to be more strong of an attractive force than negatives and this operation makes sure the aggregate strength remains unchanged