You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Brian Broom-Peltz edited this page Jul 30, 2015
·
1 revision
The scale factor is a measure of the difference in distance scales between
layers. The distance scale, dx, is the distance between neurons in
retinatopic units so dx in the retina is 1.
A column's nx and ny give the base number of neurons in each direction for
the column.
Each layer's nxScale and nyScale gives that layer's neuron density relative
to that of the column. So, if the column's nx is 32 and the layer's nxScale
is 0.5, the layer will have 16 neurons in the x-direction and dx=2.
In the OpenPV code, the scale factor is usually stored as log base 2 of the
relative scale factor. Thus if nxScale = 0.5 for a layer L, then
L->xScale == 1, and dx == 2 (the linear density of neurons is 1/2 that of
the retina). Likewise if L->xScale == -1, then dx == 1/2 (the linear
density of the neurons is 2 times that of the retina).
The scale factor is an integer. If the user-supplied nxScale is not an
integer power of two, xScale will be the integer nearest log2(nxScale).