espnet2.enh.layers.ncsnpp_utils.normalization.get_normalization

source

espnet2.enh.layers.ncsnpp_utils.normalization.get_normalization(config, conditional=False)

Obtain normalization modules from the config file.

This function retrieves the appropriate normalization layer based on the provided configuration. It supports both conditional and non-conditional normalization layers.

• Parameters:
  • config (dict) – A configuration dictionary containing model parameters. It must have the key ‘normalization’ to specify the type of normalization to use.
  • conditional (bool) – A flag indicating whether to return a conditional normalization layer. Defaults to False.
• Returns: A normalization layer class or a partial function for the specified normalization type.
• Return type: callable
• Raises:
  • NotImplementedError – If a conditional normalization type is specified that is not implemented.
  • ValueError – If the specified normalization type is unknown.

Examples

>>> config = {'model': {'normalization': 'InstanceNorm'}}
>>> normalization_layer = get_normalization(config)
>>> print(normalization_layer)  # Output: <class 'torch.nn.modules.normalization.InstanceNorm2d'>

>>> config = {'model': {'normalization': 'InstanceNorm++'}}
>>> normalization_layer = get_normalization(config, conditional=True)
>>> print(normalization_layer)  # Output: functools.partial object for ConditionalInstanceNorm2dPlus