Oribital Object Detection v2 is a complete redesign and rebuild of my Galvainze Data Science Immersive Capstone project in which I attempted to learn how to build several TensorFlow models to perform object detection on cargo ships in the San Francisco Bay from satellite imageery. I used Google Cloud Platform and targeted serverless products for hosting code where possible. The project was built specifically with a few things in mind:
- Find out if my Galvanize education was worth it's salt in the real world. If the foundations in algorithms and machine learning that I gained were truly valuable, then I should be able to pick up deep learning using a brand new framework and implement a production scale object detection model including data pipelines in 2 weeks, right?
- Create an end-to-end machine learning product, including ETL from source API through to web app for presentation.
- Be fun and an incredible challenge!
The project has been a success on all accounts! But now it's time to take it to the next phase.
This project will be focused on building a product using the same techniques and same premise, but with more cutting-edge object detection models, a focus on larger scale data processing for more than just images, consistent code testing practices, and potentially continuous integration and continuous deployment. The architecture will include micro-services separating each of the primary modules and be designed specifically to support cost effective scalability and extensibility.
So here we go!