ResNet with attention-layer on custom data set.

Here is a ResNet with attention layers that are designed for custom data sets. There is always a way for improvements, but this would get you started. The training.py is compatible with the CIFAR data sets. The attention layer is based on the following github page (commit version: 15c111d).

This is the cleaned version, maybe some mistakes namings are wrong, like the test script is missing

Usage

Install with anaconda python 3 version and Keras. Go to the train directory.
Change in vars.py the variables for your dataset.

Always double check if function load_custom_data(...) is uncommented if you learn on your own data set

# Start training
python training.py

Dataset structure

The directory names (train, validation, test) can be changed in the vars.py file if needed.
In the vars.py there are some defines that depense on your data set, like the class_names.

Dataset structure: 
        [path_to_dataset]/train/[class_directorys]/[files]
        [path_to_dataset]/validation/[class_directorys]/[files]
        [path_to_dataset]/test/[class_directorys]/[files]

Made by:

martijnvwezel@muino.nl and @RensHam