notes.txt

Data augmentation:
- makes it hard for the model to simply memorize features 
  and forces it to learn a good representation (more generalizable) of the data.
- Example: The model sees an image and decides to relate its background color with the correct class.
           Then sees an augmented image with a different color. Dang it! It has to learn something 
           different because color isn’t the one.


Improvements to do:
- Classification model without preprocessing. 
  - Without treating data imbalance. 
  - Treating data imbalance (class weights).
- Classification model with preprocessing. 
  - Without treating data imbalance. 
  - Treating data imbalance (class weights).
Compare the 2 models to see if the preprocessing has any effect. 

- Model: 
  - input_shape not specified: (224, 224, 3): the input shape of the test set matches the default assumptions made by the model.
  - Update the model architecture and specify the appropriate input shape explicitly.