Doodle Master

Inspiration

The inspiration for this app was we sometimes have trouble finding emojis just by typing their name. So we wanted the ability to find emojis by drawing them, since it is much easier to remember their shape, than their name.

What it does

Our app serves one purpose, that is it allows users to draw emojis and then see the most similar one as determined by a machine learning model.

How we built it

We built it using google collab. The actual stack was python for coding, NumPy + Pillow for the converting of drawing into model input (preprocessing), a pre-trained model by google to recognize the sketch and also our own model we trained using TensorFlow/Keras to build our own custom CNN. We also used clip to filter emojis that are too complicated for the doodles to represent and train the model with doodle data from quickdraw.

Challenges we ran into

The biggest challenge that we ran into was training time for the emoji model and ways to get around this.

Accomplishments that we're proud of

We are proud that we managed to get a working app. Even though its features are low and the model can only predict a small range of emojis it still works; to us that was a good thing.

What we learned

We learned quite a few things during this hackathon. But I think the biggest thing was how to properly work in a team. We had disagreements, but we learned how to work through them. We also gained experience on working on teams as a software project and what we should do for certain things like sharing files.