-
Front page of our website.
-
Image being dropped in the website.
-
The garbage being sorted into the correct dustbin allowing user to know where to put their trashcan.
-
Some examples to try out.
Inspiration
While wandering around the charming Charlottesville, we noticed many trash cans accompanied by multiple recycling bins. This made us rethink the sustainability and the recycling question. While doing our research, we realized that some items are recycled incorrectly or are not recycled at all, so we came up with BinGoSort.
What it does
BinGoSort is a Deep Learning model integrated into a user-friendly interface to help anyone recycle smarter. The model runs on a cloud, the user can add as many recycling categories as they want, and the model will place the object in the correct bin. Besides running on the website, one of our central ideas was to implement the model into intelligent trash cans, allowing them to automatically sort the disposed items.
How we built it
We built the model operating behind BinGoSort using a Docker container and a Virtual Machine, which are connected to a React app hosted on a domain named after the project. We deployed to the domain bingosort.tech from domain.com.
Challenges we ran into
Generally, for a project like this one, we would use a traditional classification model that could be trained in a short period of time and work with no issues. For this project, we used an NLP model. Still, we lacked data because we did not find public databases categorizing trash, which significantly elongated the time required to deploy the model.
Accomplishments that we're proud of
The UI on the React website is crispy and smooth. We used well-combined colors and Adobe Illustrator skills to make the app simple and unique. The model is accurate for the given time period, averaging 92% accuracy, which can be enhanced with training.
What we learned
We acquired many UI skills, learned how to use Docker(two of our group members had never used it before), and explored the exciting and crucial world of recycling. We also learned how to develop and run our own fast api.
What's next for BinGoSort
We will look to add sensor to the system and allow automate running. Next, we will look to increase the accuracy to an acceptable standard, prepare software for deploying the model in actual trash cans, and add more UI features for the best online experience.
Log in or sign up for Devpost to join the conversation.