Use one CNN model while trying to address common data problems: class imbalance and overfitting.
-
Manually move images around the filesystem so they are stored in a structure suitable for model training.
-
Clean up the dataset as it presents some corrupted images and other issues.
-
See how class imbalance negatively affects model evaluation.
-
Solve overfitting problems by using Data Augmentation techniques.
All of this while using the same model architecture to showcase that a lot can be achieved by doing changes to the data and not the model.