Trashify

Inspiration

About one-third of an average daily dump is made up of packaging material!

What it does

The only app that can reliably classify your trash to improve product lifestyle sustainability.

How I built it

Trashify is built with React as front-end and Flask as backend. We also incorperated Google's AutoML and trained that model to our need, in this case we taught it to distinguish the difference between 'Composte', 'Recycle', 'Trash' and 'Other'. (Other would imply that the focus of the image is not disposable i.e. humans, ceiling, etc... ).

Challenges I ran into

Several challenge did appear. Most prominent of them all is the camera. As a quick resort, we simply accessed the camera via the browser to take a quick picture which is sent to the backend. The picture taken however, if it was not focused on a single item, would contain too much information for our classification model to understand.

A solution we figured would to train our model with images with multiple items in the background and/or odd angles.

A very frustrating challenge was deployment. As easy as we thought it was, it took us many tries to attempt to deploy the application to production however setting up the server took a toll on us. At 4am our brain likely stopped working and we caved in.

Accomplishments that I'm proud of

I'm very proud of my team to being able to integrate the Google AutoML into our application. This definitely gave us the boost of confident that we required at 2am. What I learned

I learned a great deal on supervise learning and affecting it's thought process.

Lastly, I learned to hide my Google API credentials behind a private repo at the least.

What's next for Trashify

Train with more items, use another Camera library, host on a server, finally push to PWA, and incorperate software in an IoT waste bin to sort the disposed items.