This project contains 11,500 image training data of 5 common recycling items. The data was collected at Portland state University using a Nikon Camera. Images were taken at 3008x2000 resolution with JPEG basic quality. Each image contains a single item with a white colored background.
Crushed Soda Cans
We were interested in separating recycled items from trash. This can be viewed as a general anomaly detection problem where the normal class is actually five classes and anomaly class is anything else. Our work has resulted in one paper (linked above) and code that is available on GitHub.