Keras Boilerplate

A boilerplate (sort-of) for Keras Projects. I suggest using it as a boilerplate for Image related projects.

• preprocess.py - preprocessing logic to create a pre-normalized, pre-augmented, pre-regularized data. Simply put, this should prepare your raw data to an easy to work with format.
• normalize.py - normalization, augmentation, regularization logic. This process should output a format easy to feed to the model chosen for your problem
• train.py - model creation and training logic. It should use the data from the normalize process to generate a model and statistics for the visualize process such as confusion matrix, accuracy, lossn etc.
• visualize.py - create plots for better understanding of the data and training process.

Examples

The first thing I'd recommend checking.

DAL

A common name for data-access layer, meaning all data handling objects Use these tools to handle datasets and pre-processed results.

Use Dataset abstract class as an interface for implementing any other dataset objects such as Oracle / MySQL databases, or perhaps Elastic Search

Callbacks

Add custom callbacks to this package.

Metrics

Add custom metrics to this package.

Visualization

Used for visualizing the data generated either via

• Pre-processed data.
• Data augmentation.
• Normalized data.
• Model related statistics and plots.
• Model outputs.
• Post-processed model outputs.