This project was developed for the course of Artificial Neural Networks and Deep Learning for the MSc. in Mathematical Engineering at Politecnico di Milano, A.Y. 2022/2023.

.
├── README.md
├── misc
│   ├── accuracy_results.csv
│   ├── confusion.pdf
│   ├── ensemble_vgg16-resnet50.png
│   ├── single_augmentation.jpg
│   ├── supernet.pdf
│   └── ypred.npy
├── models
├── notebooks
│   ├── final_model.ipynb
│   ├── keras_tuner.ipynb
│   ├── metrics.ipynb
│   ├── supernet_choice.ipynb
│   └── supernet_histories
├── report
│   ├── bibliography.bib
│   ├── report.pdf
│   └── report.tex
├── requirements.txt
└── training_data_final

• final_model.ipynb is the main file that performs training and fine-tuning, and saves the models. Its decisions are partially based on the results from supernet_choice.ipynb and keras_tuner.ipynb. Finally, these models are evaluated in metrics.ipynb.
• keras_tuner.ipynb provides the implementation of the Keras tuner code to tune the hyperparameters of the models.
• metrics.ipynb contains the code to evaluate the models and provide accuracies and F1-scores.
• supernet_choice.ipynb performs a pilot run of transfer learning testing different pretrained models.
• accuracy_results.csv contains a table of the accuracies of some of the models that we trained during the competition, it's included for the sake of completeness and as appendix in case the report explanations wouldn't be clear enough.

• Paolo Botta (@ploki99)
• Teo Bucci (@teobucci)
• Silvia Caresana (@silviacaresana)

Check out the final report.pdf.

GNU GPLv3