We were inspired by the CSO mascot, the duck.
What it does
The model will identify ducks in images or realtime. It will also differentiate whether the duck is real or a rubber duck. You can also call or message the Twilio number and get a surprise message.
How we built it
We used the roboflow object detection model, and trained it with three-thousand annotated images of real and rubber ducks.
Challenges we ran into
We had geese in our data set, skewing it slightly. Also, it was difficult to find a large enough quality data set to train our model; as even though we trained it with three-thousand images, it still wasn't enough to get the model above eighty percent accuracy.
Accomplishments that we're proud of
We were able to successfully train a model that can find ducks live and investigate a social experiment.
What we learned
We learned about the importance of quality data when training a model, and the importance of quality annotations in order to get precise results.
What's next for QuackQuack
We would like to expand to a global scale and contribute all earnings to the "Ducks Unlimited Foundation" committed to conserving the habitats of local ducks throughout the North American continent.