Can any one help me with this question?
Lets say i have a 4x4 tensor and i want to do subsampling in the following way, so for each 2x2 block i put the smallest elements together,
then the second smallest elements, and so on, so the out put will be 4 tensors with size half of the input. Stride will be = 2.
Here is an example:
Input:
A =
[1 2 3 4
5 6 7 8
9 10 11 12
13 14 15 16 ]
Results:
A1 =
[1 3
9 11]
A2 =
[2 4
9 12]
A3 =
[5 7
13 15]
A4 =
[6 8
14 16]
And that was just a toy example, the real question is how to do it for a tensor with dimention 'BxCxMxM`