We are POT, where every (P)ieces (O)f (T)rash counts. Waste audit is an important operation. It informs people what we are throwing away everyday. For such an important task, we are still using paper and pencil to do all the manual work. Here at POT, we want to provide a solution to automate the waste audit process, make data reporting easy, and make more people involved in this waste auditing process.
What it does
POT is a platform that encourages changes for the waste. To make a change, we need data to support our statement, we need to show the effect our action, we need to work as a community to innovate for ways to tackle the waste problem in Hawaii.
POT is doing that by
1) Making data analysis easy, we developed a machine learning solution so volunteers can take a picture of the trash and upload it to our platform, where our platform does the sorting and counting. This shortens the time it takes to audit a building, allowing more data at different buildings to be collected.
2) Making data visualization easy. The public can track changes of the waste by proportion and a time series analysis to see changes throughout different dates. We want to make sure that the effort of the volunteers are credited and recognized by making the data easily accessible and understandable.
3) Making sure the community can get involved. Actions items can be created for each location, so that people that belong to the location can form their own community to take action for waste reduction. Moreover, a ranking system is in placed so people can learn which location is undergoing the the biggest decrease in waste volume. By making action items public people can also learn from communities that are doing great.
How I built it
We used Python flask to run the server, mongodb for database, PyTorch for machine learning with inceptionnetv4 to generate a set of features for the object to classify trashes, and Angular to build the front end.
Accomplishments that I'm proud of
Matt and Michele are awesome. I am proud of myself recruiting the 2. Matt did a really good job of being able to segmented out the background by doing statistics of the center and spread for the background, so that bounding boxes can be drawn around area of interest. That is probably the most technically challenged part of the project (but Matt is awesome). Michele on the other hand, hacked out the graphs quickly, that freed me up to looking into building features that allow the growth of the community.
What's next for POT (Pieces of Trash)
This project is every scalable. First public can use it in ease, hence more locations can join to form a great error; and waste classification is not pre-defined, community can choose to train on trash that they are interested in. To make it more usable, we will make the machine learning more sophisticated so it can handle more complex pictures and analysis.