https://github.com/subpic/ku

Training management utilities for TensorFlow Keras.
https://github.com/subpic/ku

deep-learning hdf5 image-processing keras keras-generators tensorflow

Last synced: 30 days ago
JSON representation

Training management utilities for TensorFlow Keras.

• Host: GitHub
• URL: https://github.com/subpic/ku
• Owner: subpic
• License: mit
• Created: 2019-08-02T07:42:16.000Z (over 5 years ago)
• Default Branch: stable
• Last Pushed: 2022-10-31T13:31:25.000Z (about 2 years ago)
• Last Synced: 2024-09-28T09:09:32.173Z (4 months ago)
• Topics: deep-learning, hdf5, image-processing, keras, keras-generators, tensorflow
• Language: Python
• Homepage:
• Size: 673 KB
• Stars: 1
• Watchers: 4
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Keras Utilities (KU)

The project contains utilities for image assessment development with Keras/Tensorflow, including utilities for model training, custom generators, image management and augmentation. This is an extension of [kutils](https://github.com/subpic/kutils).

## Overview

Some of the key components of each file:

**`model_helper.py`**:

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

**`generators.py`**:

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

**`image_utils.py`**:

* various utility functions for manipulating images (read, write to HDF5, batch resize, view batch)

**`image_augmenter.py`**:

* `ImageAugmenter`: Create custom image augmentation functions for training Keras models.

**`generic.py`**:

* `H5Helper`: Manage named data sets in HDF5 files, for us in Keras generators.

* `ShortNameBuilder`: Utility for building short (file) names that contain multiple parameters.

**`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

You can find more information in the docstrings.