Normalization

USAGE

All functions can be imported using layers module

2D Layers

normalization_2d(x, axes=(1, 2), epsilon=1e-08, name='norm_layer_2d')

Apply a zero mean - unit variance normalization (standardization)

Parameters:

• x – 4D Tensor to normalize. It expects a data in format NHWC.
• axes – Axes where apply the normalization. Default: (1, 2) - [Height, Width].
• epsilon – A small floating point number to avoid dividing by zero.
• name – Operation name.

Returns: Normalized tensor

3D Layers

normalization_3d(x, axes=(2, 3), epsilon=1e-08, name='norm_layer_3d')

Apply a zero mean - unit variance normalization (standardization)

Parameters:

• x – 5D Tensor to normalize. It expects a data in format NDHWC.
• axes – Axes where apply the normalization. Default: (2, 3) - [Height, Width].
• epsilon – A small floating point number to avoid dividing by zero.
• name – Operation name.

Returns: Normalized tensor