Annotation services

Semantic segmentation annotation services

Semantic segmentation annotation assigns a class to every pixel in an image — road, sidewalk, vegetation, sky — producing a full-frame mask. Unlike instance segmentation it does not separate individual objects of the same class; it describes what every part of the scene is.

Trains
Semantic segmentation
From
$0.18 / label
Review
Every job
Formats
PNG class-index masks

Is it the right label?

When to use semantic segmentation

Semantic segmentation — the model labels every pixel of a new image, for scene understanding and area measurement.

  • The question is "what is everywhere", not "where are the objects": drivable area, land cover, field coverage.

  • Classes are regions, not countable things: sky, water, soil, background tissue.

  • You need area or coverage percentages per class across the whole frame.

When something else fits better

If you need to count or track individual objects, semantic masks merge touching instances into one region. Use instance (polygon) segmentation, or combine both as panoptic labeling.

Polygons & instance segmentation

How we label it

Our process

  1. 01

    Define the class map

    A complete, ordered class list with an explicit background or void class, and a rule for every boundary between two classes.

  2. 02

    Region-first labeling

    Large regions are filled first with assisted tools, then boundaries are refined by hand so no pixel is left unassigned.

  3. 03

    Boundary refinement

    Edges between classes — where models fail — are labeled at full zoom, following the guideline's tie-break rules.

  4. 04

    Full-frame review

    Reviewers check coverage (no holes, no stray pixels) and class confusions before the frame is approved.

Quality

What our QA checks

Every job is reviewed by a person other than the annotator, on top of automated checks. Work below the agreed threshold goes back for rework.

Per-class IoU against reviewer-corrected masks

Unlabeled-pixel and hole detection

Boundary consistency between adjacent classes

Class palette and index checks on every exported mask

Pricing & delivery

Rates and formats

AnnotationRate
Semantic segmentation$0.18 / label

QA review is included. Bulk discounts apply to large projects; the final quote is confirmed after we review a sample of your data. Full pricing

Delivered as

PNG class-index masksCOCO JSON (RLE)Cityscapes-style masksCVAT XMLDatumaro

Used for

Autonomous driving scene parsing · Satellite and aerial land cover · Medical tissue segmentation · Agricultural field mapping · Industrial surface inspection

Semantic segmentation FAQ

Semantic segmentation, answered

What is semantic segmentation annotation?

It is labeling every pixel of an image with a class, such as road, car or sky, to produce a full-frame mask. Models trained on it understand the whole scene rather than detecting individual objects.

What is the difference between semantic and instance segmentation?

Semantic segmentation gives every pixel a class but treats all objects of that class as one region. Instance segmentation separates each object, so two touching cars are two labels. Panoptic segmentation combines both.

Why is semantic segmentation more expensive than bounding boxes?

Every pixel must be assigned and every class boundary drawn precisely, which takes several times longer per image than boxing a few objects. Assisted tools reduce the time, but boundary refinement stays manual.

What format are semantic masks delivered in?

Usually PNG images where each pixel value is a class index, plus a class map file. COCO RLE, Cityscapes-style masks and CVAT XML are also available.

Need semantic segmentation on your data?

Send us a few sample images or clips. We'll agree the guideline with you and send a plan and a per-label quote.