lukalake / kutils

Keras extended utilities

Keras extended Utilities for MLSP feature learning

The project contains tools for easier development with Keras/Tensorflow. It was developed for AVA-MLSP feature learning. The code in applications.py and model_helper.py pertains more to MLSP feature learning, wheras the remaining tools are general purpose.

Overview

Some of the key components of each file:

generic.py:

• H5Helper: Manage named data sets in HDF5 files, for us in Keras generators.
• pretty: Pretty-print dictionary type objects.
• ShortNameBuilder: Utility for building short (file) names that contain multiple parameters.

image_utils.py:

• ImageAugmenter: Create custom image augmentation functions for training Keras models.
• read_image, read_image_batch: utility functions for manipulating images.

model_helper.py:

• ModelHelper: Wrapper class that simplifies default usage of Keras for regression models.

applications.py:

• model_inception_multigap, model_inceptionresnet_multigap: Model definitions for extracting MLSP narrow features
• model_inception_pooled, model_inceptionresnet_pooled: Model definitions for extracting MLSP wide features

generators.py:

• DataGeneratorDisk, DataGeneratorHDF5: Keras generators for on-disk images, and HDF5 stored features/images

You can find more information in the Python Docstrings.