Create a script training a CNN using the created dataset

[jkbdnj]

Task description

Create a script that would use to train a CNN to classify diseases from plant leaf images. The important things to consider are:

• choosing the right CNN (based on number of parameters within the CNN, number of pictures in the dataset, regularization method used),
• modification of the CNN model for the problem settings (number of output units as the classes in the dataset),
• dataset division between training, testing and validation subsets,
• dataset augmentation,
• training results that should be logged,
• saving the training results in a way that a python library may be used to create the charts.

[jkbdnj]

NOTE: When selecting a CNN architecture, I am also limited by the architectures provided by keras that are pre-trained on Imagenet dataset: https://keras.io/api/applications/.

The most promising CNN model is the: EfficientNetB0.

[jkbdnj]

NOTE: Typical transfer learning workflow + fine-tuning: https://keras.io/guides/transfer_learning/#the-typical-transferlearning-workflow.

There is also no need to resize the images in the datset because they have already been resized oto 256 256 pixels. The EfficientNetB0 has the spatial input dimensionality of 256 256 pixels.