Home / Label types / Image

Semantic and instance segmentation

Every pixel in the image assigned to a class, or to a single object when you need to tell instances apart.

Street image 0144QA passed
skybuildingvegetationroadcarperson
AnnotatorLabels drawn
ReviewerChecked
QAApproved

What we label

  • Semantic masks for whole scenes
  • Instance masks for separate objects
  • Background and ignore regions
  • Class colour maps that match your spec

Where it is used

  • Scene understanding for robots
  • Road and street segmentation
  • Indoor rooms and furniture
  • Product cut-outs for e-commerce

How we check quality

  • Mask edges checked at full resolution
  • No unlabelled pixels left behind
  • Overlaps between objects resolved
  • Class colours verified against the legend

Delivery formats

We deliver in the format your training pipeline already uses. Need something else? Send us a sample file and we will match it.

PNG masksCOCO RLECVAT XMLCityscapes-style

See segmentation on your own data

Send a small sample and your guidelines. The pilot is free.

Request a free pilot