Documentation

BatchNorm

Input

  • x:
    • Required: yes
    • Type: float16, float32
    • Description: input tensor
    • Restrictions: none
  • mean:
    • Required: yes
    • Type: float16, float32
    • Description: mean
    • Restrictions: none
  • variance:
    • Required: yes
    • Type: float16, float32
    • Description: variance
    • Restrictions: none
  • scale_factor:
    • Required: yes
    • Type: float16, float32
    • Description: decay value for the moving mean and variance computation.
    • Restrictions: none
  • scale:
    • Required: no
    • Type: float16, float32
    • Description: reserved
    • Restrictions: none
  • offset:
    • Required: no
    • Type: float16, float32
    • Description: reserved
    • Restrictions: none

Attribute

  • eps:
    • Required: no
    • Type: float
    • Description: a small number to avoid division by zero while normalizing (default = 1e – 5)
    • Restriction: The value is user-defined.
  • use_global_stats:
    • Required: no
    • Type: bool
    • Description: If True, uses accumulated mean and variance values for normalization. Fixed at True.
    • Restriction: Must be True.

Output

y:
  • Required: no
  • Type: float16, float32
  • Description: output tensor
  • Restriction: Must be of type float16.