poke-cans

• 

  Testing a trained model

Inspiration

We felt inspired by games like Pokémon Go, and how that got kids out and about more, showing that a fun application can get a young child out and interested in something good for them, or in this case, the environment.

We hope the project will encourage younger children to go out collecting rubbish, looking for new cans to collect!

What it does

When a user logs in, they can either upload a photo or view their collection (or, as an added Easter egg, "make their day greener ;)". Once a photo is uploaded, AI recognition is used to determine the brand of the can and then add it to your collection, in which you can see every can that you have recycled, as well as their "stats" (sort of like Pokémon), which detail useful bits of information about the drink such as the price and a general healthiness rating.

How we built it

Coffee and late nights! We decomposed the project into smaller pieces for each of our team members to handle, who then worked on their part of the project, until we came together to combine them into the final project on Saturday.

Challenges we ran into

There was an issue with the system we used to generate the training data, causing all the data to use "blue charge" cans, meaning we at first struggled to get the AI working correctly. We fixed this as a team. Also, we had issues with training the AI on our machines, so we ended up using google colab instead, which worked much better, except for when it deleted the model we trained for 200 epochs. The time period was so much shorter than anything that we were used to, we struggled to put the final project together. we had each part of the project working individually, just unfortunately not together. We faced many other issues too, but working as a team, working hard and our experience helped us overcome those issues with ease. Furthermore, we were not able to put together a presentation video in time, so I hope you appreciate the rickroll :)

Accomplishments that we're proud of

We were really proud of the entire project and how it turned out because it was our first time ever working on a project like this, and prior to this we did not have much experience in web design and limited-time coding. We were especially proud of the can classification algorithm, as this was the first time some of our team members had interacted with

What we learned

I, personally, learned about Siamese networks and their applications. My team member Maxime learned how to use HTML, CSS and JavaScript in front-end design, as well as learning about several of the Python modules used in the web back-end My team member Lucy learned about the basics of training an AI as well as the full process of developing a program from start to (albeit rushed) finish. We believe all of these skills may be useful for future coding projects as a team, and for future hackathons. We had a lot of fun, doing our first, and certainly not last, hackathon!

What's next for poke-cans

We will likely continue developing it, possibly into a mobile application, but since none of us have experience with that, only time will tell!

Built With

• flask
• github
• html5
• keras
• python
• replit
• tensorflow