文档

ArgMax

Input

x:
  • Required: yes
  • Type: fp32, fp16
  • Description: input tensor
  • Restriction: none

Attribute

  • axis:
    • Required: no
    • Type: int
    • Description: axis to be reduced in the input tensor. If this parameter is not provided, top_k is calculated per batch.
    • Restriction: none
  • out_max_val:
    • Required: no
    • Type: bool
    • Description: whether to output the maximum value.
      • If out_max_val is True and there are two top blobs:
        • If axis is specified, only the maximum value of each axis is output.
        • If axis is not specified, the index of the maximum value and the maximum value are output.
        layer {
  name: "argmax"
  type: "ArgMax"
  bottom: "data"
  top: "indices"
  top: "values"
  argmax_param {
    out_max_val: True
    top_k: 1
  }
      • If out_max_val is True and there is one top, the maximum value is output.
        layer {
  name: "argmax"
  type: "ArgMax"
  bottom: "data"
  top: "values"
  argmax_param {
    out_max_val: True
    top_k: 1
    axis: 1
  }
}
      • If out_max_val is False, the maximum index of each axis is output.
        layer {
  name: "argmax"
  type: "ArgMax"
  bottom: "data"
  top: "indices"
  argmax_param {
    out_max_val: False
    top_k: 1
    axis: 1
  }
}
    • Restriction: none
  • top_k:
    • Required: no
    • Type: int
    • Description: defaults to 1, indicating the top_k in each axis. The value range is [1, x.shape(axis)], corresponding to top_k in Caffe.
    • Restriction: Must be 1.

Output

  • indices:
    • Required: no
    • Type: int32
    • Description: index of the maximum output value
    • Restriction: none
  • values:
    • Required: no
    • Type: fp32, fp16
    • Description: output tensor, including the maximum value index or maximum value
    • Restriction: none