BatchNormalization

Description

Standard tensor.

Input

Five inputs:

X: 4Dtensor of type float16 or float.

scale: (optional) tensor of type float, specifying the scale factor.

B: (optional) tensor of type float, specifying the offset.

mean: (optional) tensor of type float, specifying the mean value.

var: (optional) tensor of type float, specifying the variance.

Output

Five outputs:

Y: normalized tensor of type float16 or float.

mean: (optional) tensor of type float, indicating the mean value.

var: (optional) tensor of type float, variance.

saved_mean: (optional) tensor of type float. During training, the saved mean value is used to accelerate gradient calculation. (Not supported in versions later than v14.)

saved_var: (optional) tensor of type float. During training, the saved variance is used to accelerate gradient calculation. (Not supported in versions later than v14.)

Attribute

epsilon: (optional) float, added to var to avoid dividing by zero. Defaults to 0.0001.

momentum: float. This parameter is not supported currently.

ONNX Opset Support

Opset v8/v9/v10/v11/v12/v13/v14/v15/v16/v17/v18