Documentation

Normalize

Input

  • x1:
    • Required: yes
    • Type: float32, float16
    • Description: input tensor to be normalized
    • Restrictions: none
  • x2:
    • Required: no
    • Type: float32, float16
    • Description: scale tensor, ND

      If channel_shared==True, then N=1. If channel_shared==False, then N=channel.

    • Restrictions: none

Attribute

  • across_spatial:
    • Required: no
    • Type: bool
    • Description: If True (default), normalizes CHW. If False, normalizes C only.
    • Restriction: either True or False
  • channel_shared:
    • Required: no
    • Type: bool
    • Description: If True (default), shares x2 across channels.
    • Restriction: either True or False
  • eps:
    • Required: no
    • Type: none
    • Description: a small number to avoid division by zero while normalizing (default = 1e – 10)
    • Restrictions: none

Output

y:
  • Required: yes
  • Type: float32, float16
  • Description: output tensor
  • Restrictions: none