Waste & recycling

Material, contamination and condition, labelled on fast moving streams.

Annotation and live review for smart bins and sorting lines, including construction waste.

From the bin to the bale

Every part of the workflow where we can take work off your team.

Show
Bin fill level
Contamination at the kerb
Bin camera footage collected

Collection

Material type on the belt
Dense stream segmentation
Robot pick targets

Sorting

Debris type
Recoverable material
Skip contents

Construction waste

Review decisions logged
Sorts verified live
Purity estimates checked

Compliance

We work with conveyor video, bin cameras and images.

What we label

Every class is written into a guide with your team before the first batch. Files come back as COCO JSON, YOLO, Pascal VOC, CSV, JSONL or your own schema.

Conveyor and bin video

  • Items

    Instance mask

    PET, HDPE, cardboard, aluminium, glass

  • Contamination

    Class and bounding box

    Food residue, film, hazardous item

  • Condition

    Class

    Crushed, wet, mixed, clean

Images

  • Construction debris

    Polygon and class

    Concrete, timber, metal, plasterboard

  • Fill level

    Class

    Empty, half, full, overflowing

How we test the model

The tests that matter for waste & recycling models, and what you get back from each.

Purity estimates

Model purity estimates compared with reviewed samples.

You get backError by material and line

Pick target accuracy

Robot pick targets checked on sampled frames.

You get backMissed and wrong picks

Audit sampling

Logged decisions resampled for compliance.

You get backAudit trail, disagreement log

Where waste & recycling teams start

Problems teams bring us, and the first piece of work we would scope.

Items overlap on a fast belt and the model sees one blob.

What we doInstance masks on dense streams, item by item.Annotate

You need proof of what was sorted, not just a number.

What we doEvery review decision logged for your compliance reports.Live review

Bring us the waste & recycling data your model gets wrong.

Send a small batch and get it back done, free, usually within 24 to 48 hours. You judge the work before you pay for anything.