Inspiration
After watching a seminar my school held, I got the idea for this. There was a discussion about how much plastic goes into the ocean. The fact that not much has been done about this problem and that it is getting worse every day is shocking to me. Having watched that seminar, I came to the conclusion that I needed to contribute. Around the same time, Mr. Beast created #TeamSeas, an initiative to clean up the seas. Video clips from their video show a machine moving through rivers and cleaning up plastic flowing into the sea. As a result, I came up with the idea of making a plastic detector as part of another idea. And then Launchology vision gave me the opportunity to build it.
What it does
For now, it basically detects 5 classes: Bottles, bags, plates, containers, and cups which make up most of the plastic waste in the ocean. It is the most accurate for bags and bottles.
How we built it
I built it using the algorithm YOLOv5 and Pytorch. Trained it in Google collab to know the perfect weights.
Challenges we ran into
I ran into a lot of challenges, before I came across YOLOv5, I was trying to install YOLOv4 (Darknet) following the videos on youtube but had some issues in installation which I could not resolve, also tried TensorFlow's SSD models but even that had issues. After all this, I was exhausted and thought of giving up but that's when I came across YOLOv5 which resolved everything.
Accomplishments that we're proud of
So this is just a prototype of what I will originally be building with much more data to be trained on. But it still worked pretty well, being a prototype. It proved that making a plastic object detector is possible.
What we learned
I learned a lot about neural networks and also learned a lesson, that there is a solution to every problem you are facing right now and you will eventually come across it.
What's next for Plastic Object detector
Next, I will just try to improvise the performance of the detector and try my best to make it useful.
Log in or sign up for Devpost to join the conversation.