Recycling and environment
Instance segmentation and QA for waste-stream classification, at production volume.
Who it is for
Recycling, waste-sorting and environmental AI teams building models that recognise materials on conveyors and in tipping or dump areas.
Typical problems
Instance labels on conveyor or dump imagery across paper, cardboard, Tetra and other classes. Materials overlap, deform and arrive in mixed or hybrid forms, so boundaries and class rules have to be precise and applied the same way in every batch.
What we can do
Segmentation and QA for waste-stream classification at production volume: instance masks, hybrid-class rules and sampled error review, delivered in regular batches.
Proof
Our waste stream classification work for a confidential client in the Netherlands: about two years of monthly batches and roughly 16,800 objects per batch in audited samples.
Read the original published waste-management story from the Netherlands, covering the 2022–23 project.
Delivery, security and quote
Every project runs through the same method: an agreed scope and taxonomy, calibration where the project calls for it, production by trained annotators and review before anything is handed back. Read more in How We Work and Tools and Delivery.
Data handling, including access, transfer, retention and deletion, is agreed with you in writing before work starts. See Security and Compliance for how we handle data.