caffe CNN: Pooling across the channels

0
votes

I have a nxmx16x1 conv layer and I would like to do pooling across the channel, so the result has dimension of nxmx1x1.

any suggestions? as far as I know pooling does not have axis param, right?

1
neural-networkdeep-learningcaffeconv-neural-networkpycaffe

1 Answers

1
votes

You don't need axis param. Simply use non-uniform kernel/stride:

layer {
  name: "pool16x1"
  type: "Pool"
  bottom: "input"
  top: "output"
  pooling_param {
    kernel_size: 16
    kernel_size: 1
    stride: 16
    stride: 1
    # ...
   }
}

should do the trick...