In 2020, when everyone was forced to self-isolate at home, we invited users to photo and label the most popular packages. Through environmental crowdsourcing, we managed to collect a giant database of marked waste and train the neural network.
Labeling was made with dounding box on images inside TrashBack mobile application. Then all lebels were validated by our employee. We pay Ecoins for valid markup
The tagged data is used to train a convolutional neural network that finds waste in the image, identifies it and localizes it
For the neural network to be of practical use, we have designed and manufactured devices that accept waste from residents, recognize it, and pay Ecoins for the recyclable materials they have returned.
packaging types were selected based on popularity, recyclability and recyclability
more than 800 people regularly mark waste in the app
users photographed the packaging before throwing it away