Garbify

Main UI

Inspiration

Most people are uninterested in classifying their trash but maybe one day this process can be made autonomous with the help of our ML model.

What it does

In order to accomplish this we created an application that allows your camera to classify the recycling items. Aside from classifying the object for you, you can also compete with your friends by classifying different examples of trash to be recycled. Friendly competition is introduced through its leaderboard and the different examples are pulled from a dataset. Through repetitive examples you can learn which items belong in the trash bin and which belong to the recycling bin. Lastly through this constant feedback we can help introduce new images into the dataset that are hard to classify and have the help of its users improve the model.

How we built it

The core classification model is built from ResNet-50, a pretrained Convolutional Neural Network, trained with more than 2000 pictures of recycle and trash. Our current deep learning model achieves higher than 99% accuracy. Also, we could collect more data under users’ permission and perfect the model then.

Challenges we ran into

We ran into challenges learning Flutter, which we had little experience with and had to learn essentially from scratch. It was difficult coordinating across time zones, with members of our group located in Pacific standard time, Eastern standard time, and Indian standard time.

Accomplishments that we're proud of

Machine learning integration with the built in camera was something that we were really proud of and managed to successfully finish.

Implementing the authentication system properly and being able to have a scalable system for players all around the world to enjoy, as well as having a diverse team of developers around the world.