Given the problem statement I wanted to make a solution built to take advantage of GPU processing power and parallelization ideally to create a model that is permanent and scalable.
What it does
Aiming to create the Highest Performing Solution by using parallel computing techniques. Dire Drone uses APIs from Nvidia's Rapids Opensource libraries to efficiently do computations exclusively on the GPU. to compare my solution I also implemented a test client that uses Microsoft cognitive services to check if it quickly and accurately detect a fire in a scene better than the cognitive services api.
How I built it
Initial environment setup was done with the getting started Repo on MS Azure Notebooks, where credentials and the Fire Drone scenes were created with the FireDrone SDK. I followed a format similar to this project (https://medium.com/the-artificial-impostor/umap-on-rapids-15x-speedup-f4eabfbdd978) in order to learn how to use Rapids (https://rapids.ai/about.html) to create to create a model to detect fires. All of the SDKs, APIs, and Libraries are Python 3 based.
Challenges I ran into
Getting setup with the Fire Drone SDK and understanding how the Drone interacted with the simulated environment took some time to understand.
Accomplishments that I'm proud of
Happy to have contributed a solution to a project that focuses on AI for Good. I enjoyed interacting with the FireDrone.AI community on Slack and using the tools given to come up with a unique solution along with all of the other participants.
What I learned
I gained a lot of hands on Data Science experience by participating in this challenge and researching and implementing different solutions as I went along.
What's next for Dire Drone
Doing performance benchmarks and working with the Rapids community to learn more about how I can use the APIs to there full extent to improve my solution. I plan to deploy my solution to a Tello Drone.