Inspiration
We've all seen cars on the road that looked super cool but it can get hard identifying them. That's why we decided to make CarDar.
What it does
Users can submit photos of cars and CarDar will segment the image into the main car that is in shot. It will then pass it to a CNN that we can load through a dumped pickle file after it is trained to run a .predict() on that segmented image to get a result of what the car is.
How we built it
We used, Python, OpenCV, Flask, PyTorch and Stanford's Car AI Dataset
Challenges we ran into
It was hard to find a well-working version of OpenCV's haarcascade
Accomplishments that we're proud of
We really like the idea and would definitely use it on our own time!
What we learned
None of use had experience in deep learning, only in traditional ML, so it was a huge learning opportunity for us using both OpenCV and PyTorch
What's next for CarDar
Getting some implementation with Google Maps that shows nearest dealerships for these cars
Log in or sign up for Devpost to join the conversation.