PoC for the Unimi laboratory "Deploy Machine Learning Models on Google Cloud Platform" held by Emanuele Guidotti. The project uses a self-trained Keras CNN, served by a quick and dirty Streamlit backend, to classify images uploaded by the user. The app is deployed publicly through a Docker Image on a container served by Google Cloud Run.
Dataset used for training: https://www.kaggle.com/datasets/alessiocorrado99/animals10 (10 animals)
A possible improvement was only conceptually explored due to the one-day time constraint the team set.
In this improvement the problem of monitoring the continuous performance of the model on user input gets addressed;
we added a CloudSQL database, on which inputs and predictions are saved to be assessed in a later moment.
