In this chapter, we have implemented a complete real-life production, from training to serving a deep learning model. We also created a web interface in a Flask app so that users can upload their images and receive results. Our model can automatically be fine-tuned every day to improve the quality of the system. There are a few things that you can consider to improve the overall system:
- The model and checkpoints should be saved in cloud storage.
- The Flask app and TensorFlow Serving should be managed by another, better process management system, such as Supervisor.
- There should be a web interface so that the team can approve the labels that users select. We shouldn't rely completely on users to decide the training set.
- TensorFlow Serving should be built with GPU support to achieve the best performance.
