In 2022, a Dutch client approached us with a task involving labelling different waste items into 26 pre-defined classes including Paper, Metal, Plastic, PVC, Brown cardboard, Tetra, PET bottle, Aluminum foil, etc. The goal was to optimize their AI model for better sorting efficiency. Among multiple vendors they spoke to, we’re proud that they partnered with us, and we ended up delivering over 1 million labels across 15k+ images.


 

Year: 2022-23
 

Tool Used: Supervisely
 

Challenges We Faced: 
This project required our annotators (and subsequently reviewers!) to keep a very keen eye since so many items in low resolution garbage photos are very difficult to differentiate. For example, “Paper” included all non-cardboard items except small shreds, which required precise differentiation. Similarly, the 'Other' category encompassed a variety of items such as clothes, plastics, styrofoam, and rubber – adding to the complexity.

Despite this challenge, we were able to successfully deliver the labelling work with a high degree of quality & accuracy thanks to the dedication of the special team of annotators that we were able to assemble and train. Another win in the bag for Abelling!