The mesonet package provides functions to build, train, obtain predictions from and visualize MesoNet variants.

It has been divided into four main modules:

• model - This module provides functions to build and obtain predictions from MesoNet variants. It also consists of helper functions such as obtaining a classification report, saving the model to a file/directory, saving model history, etc.
• data - This module provides functions to apply augmentations and load images from the dataset for training and testing.
• train - This module provides a single function to train a MesoNet variant.
• visualization - This module provides functions to visualize loss curves and the intermediate convolutional layers.

There is also a utils module for the purpose of storing any miscellaneous utilities required in the package. Currently, it only stores the input size of images as the constant IMG_WIDTH. The default value is 256. Therefore, by default, any model you train will use 256 x 256 images as its input. You can easily change this by modifying the constant.