Pictonet (AI Pictionary)

The home page where players can create and join games.
The game screen showing the drawing prompt and the leader board.
The result screen which shows the top three players.

Inspiration

We were inspired by Google Quick Draw and Skribbl.io to create a project that combines machine learning and online web games.

What it does

Players connect to our server and create private games which they can share with their friends using a code. Each game has 10 rounds in which every player is given the same randomly-selected prompt and must draw it as quickly as possible. If the convolutional neural network successfully categorises their drawing, they gain points which go on a leader board.

How we built it

We used C# to program the backend with SignalR for low latency messaging between the backend and the frontend. We wrote the front end using HTML, CSS and JavaScript. We created the convolutional neural network in Python using Tensorflow and exported it to JavaScript. To train it, we used the Google Quick Draw dataset.

Challenges we ran into

The neural network posed the greatest challenge as it was important that it performed well during the game. Also, getting the image data from an HTML canvas and passing it to the neural network in a specific format was difficult.

Accomplishments that we're proud of

We're proud of the look of the UI because we think it fits our game well. We're also proud of the accuracy of our neural network and the website's backend.

What we learned

We learned how to use Tensorflow and better work as a team using Git. We also learned more about web development and UI design.

What's next for Pictonet (AI Pictionary)

We'd like to add a way to configure the game (e.g., select number of rounds and difficulty) and add more drawing prompts that the neural network can learn to recognise.