Index

Preprocessing Layers API

This module provides a variety of preprocessing/augmentation layers to build custom tf.data.Dataset pipelines. NSE provides layers for both 1D and 2D input data and doesnt assume 2D input data to be images. In addition, all layers inherit from BaseAugmentation and TFDataLayer. These two layers provide the following functionalities:

• Dynamically set backend to TensorFlow for pipeline layers
• Coerce input data to have batch dimension and converted to nested dictionary
• Output data will revert to original format (e.g. no batch)
• By supporting nested dictionary, it allows layers to manipulate labels
• The layers map either sequentially or in parallel across the batch dimension

Classes:

• AmplitudeWarp –

  Amplitude warping layer

• AugmentationPipeline –

  Augmentation pipeline

• BaseAugmentation –

  Base augmentation

• BaseAugmentation1D –

  Base 1D augmentation

• BaseAugmentation2D –

  Base 2D augmentation

• CascadedBiquadFilter –

  Cascaded biquad filter

• FirFilter –

  FIR filter

• LayerNormalization1D –

  Layer normalization 1D

• LayerNormalization2D –

  Layer normalization 2D

• RandomAugmentation1DPipeline –

  Random augmentation 1D pipeline

• RandomBackgroundNoises1D –

  Random background noises 1D

• RandomChannel –

  Random channel

• RandomChoice –

  Random choice

• RandomCrop1D –

  Random crop 1D

• RandomCrop2D –

  Random crop 2D

• RandomCutout1D –

  Random cutout 1D

• RandomCutout2D –

  Random cutout 2D

• RandomFlip2D –

  Random flip 2D

• RandomGaussianNoise1D –

  Random Gaussian noise 1D

• RandomNoiseDistortion1D –

  Random noise distortion 1D

• RandomSineWave –

  Random sine wave

• Resizing1D –

  Resizing 1D

• Resizing2D –

  Resizing 2D

• Rescaling1D –

  Rescaling 1D

• Rescaling2D –

  Rescaling 2D

• AddSineWave –

  Add sine wave

• SpecAugment2D –

  SpecAugment 2D

Classes