Inspiration
We wanted to reduce carbon emission while gamifying garbage sorting.
What it does
Users would scan an item, which would be categorized into one of the four following categories: organic, paper, containers, garbage. From here, the user would be prompted to choose the correct category, after which they would receive in-game points if they selected the correct category. Users can collect these points to eventually "purchase" a tree insofar as it would be planted somewhere in the world.
How we built it
Our team was split into the frontend and backend developers. The frontend developers used figma to design the iOS application into a unique looking template with a clean interface. The colours convey a sense of environmental friendliness, and the overall app design makes it very intuitive to use. On the backend, our developers worked to create an object recognition application using AI. We wanted to have an AI application that would recognize the object it sees through the camera, and place it in one of the four categories. The categories being: organic, paper, containers and garbage. We built the application in python and used frameworks such as pytorch to implement our object recognition AI.
Challenges we ran into
We initially tried to go through 10 epochs of image training, which was too time-consuming for our current hardware given the time-constraint of the Hackathon. Although we did some image training, we had to work with the epochs that were finished rather than using all 10 epochs-worth.
Accomplishments that we're proud of
We were able to set up a demo of our project which would recognize certain items and categorize them with relative accuracy (items that were not trained within our AI may produce sporadic results). Additionally, we created a sample front-end design that would showcase what the user would see.
What we learned
For the backend developers, we learned a lot about how to set up machine learning to go through epoches to train the AI to recognize images. It was an intriguing process to learn about training loss and validation loss, as well as what results are significant to help us better train the AI.
What's next for EcoTrack
For the backend, we would want to progress through image training our AI to increase its accuracy in recognizing certain items. For our frontend, we would like to incorporate more features and a smoother interface design.
Log in or sign up for Devpost to join the conversation.