Inspiration

It's always good to be up to date with computer vision benchmarks now that computers are more than capable of dealing with simpler problems.

What it does

Classifies drawn images into 15 different classes.

How we built it

We created a DNN that we hyper parameter tuned with the help of Gaussian Processes via Bayesian Optimization.

Challenges we ran into

The data is not well behaved and requires preprocessing.

Accomplishments that we're proud of

We achieved the first place in the scoreboard.

What we learned

Computer vision applied to benchmark problems

What's next for Pictionary Pugle

Adding more classes, and more robust validation schemes.

Built With

Share this project:

Updates