Besides reducing the cost of computing, is there any other pros or cons for separating filters?
amilich
^ Separating filters basically just means representing them as 1D convolutions. So, you can only represent some filters in this manner (because they essentially require the matrix to be rank one). You lose more complex filters but gain computing power and the 1D representation.
Besides reducing the cost of computing, is there any other pros or cons for separating filters?
^ Separating filters basically just means representing them as 1D convolutions. So, you can only represent some filters in this manner (because they essentially require the matrix to be rank one). You lose more complex filters but gain computing power and the 1D representation.