eltwise sum over multiple channels

[Moti Agrawal]

Does anybody know if eltwise sum is define over multiple channels.
In other words if my bottom layer has 16 channels. Can i do eltwise sum in effect summing all 16 channels

//produces 16 channels
layers {
  name: "conv_1"
  type: CONVOLUTION
  blobs_lr: 1.
  blobs_lr: 2.
  convolution_param {
    num_output: 16
    kernelsize: 1
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
    }
  }
  bottom: "data_1"
  top: "conv_1"
}


//sum over 16 channels
layers {
  name: "sum_1"
  type: ELTWISE
  bottom: "conv_1"
  top: "sum_1"
  eltwise_param {
    op: SUM
    coeff: 1
  }
}

[lei zou]

hi, i wondered the answer too. have you got the right answer? best wishes!

在 2014年11月17日星期一 UTC+8上午7:25:49，Moti Agrawal写道：

[Beanfrog Green]

Maybe you can try SLICE the blobs to 16 blobs across the channels, then use ELTWISE SUM to sum the 16 blobs

在 2014年11月17日星期一 UTC+8上午7:25:49，Moti Agrawal写道：

Does anybody know if eltwise sum is define over multiple channels.

[Beanfrog Green]

Again, see https://groups.google.com/d/msg/caffe-users/M5japI--dAw/3G-Jc8WafzgJ
we can use 1x1 conv to do sum cross maps, here is a example:

//sum over 16 channels
layers {
  name: "sum_1"

  type: CONVOLUTION
  bottom: "conv_1"
  top: "sum_1"
  blobs_lr: 0
  convolution_param {
    num_output: 1
    bias_term: false
    kernel_size: 1
    weight_filler {
      type: 'constant'
      value: 1

    }
  }
}




在 2014年11月17日星期一 UTC+8上午7:25:49，Moti Agrawal写道：

Does anybody know if eltwise sum is define over multiple channels.

[Teddy Ma]

It's really a good idea,thx a lot!

在 2015年3月30日星期一 UTC+8下午3:39:44，Beanfrog Green写道：