Inspiration: Keeping it simple!
What it does: Uses Microsoft Custom Vision API to detect if there is fire in the current image.
How I built it: Using Microsoft Custom Vision API and Python. The Drone is moved to top-left position of screen. Starting from here, the drone moves and scans the entire scene, evaluating every frame at each position in the process.
Challenges I ran into: I had difficulty figuring out the right way to use the native Python API for custom vision. I was blocked and eventually gave up on this approach. I was able to use Postman to test the endpoints for prediction. I used Postman code generation feature to convert the Postman request to Python code which unblocked the project.
Accomplishments that I'm proud of: Integrate Microsoft Custom Vision API, Python and Postman
What I learned: How to use custom vision services offered by Microsoft.
What's next for Microsoft-FireDrone-AI: The current training set only has 129 positive and 184 negative images. The images are diverse but still the total number of images is significantly small in terms of todays deeplearning standard. Next steps would be to increase the training images to build a more robust model.