Polygon Annotation
Services

We label still images and video frames with polygon masks, including irregular shapes, overlapping objects and dense scenes. Labeling can be instance-level or class-level, using a schema built around your model's requirements. This is one of the 15+ annotation types in our computer vision data labeling services.

Vertex placement follows client-defined annotation guidelines specifying exactly how object boundaries should be traced, including edge snapping rules, vertex density per object class and handling of curved vs. angular contours. Annotators account for occlusion, lighting and depth context. mIoU scoring and vertex checks during QA confirm the rules are applied consistently across every batch.

Occluded objects are annotated by tracing the visible boundary only, unless your guidelines ask for the hidden contour to be estimated as well. Annotators follow per-class occlusion handling rules, flagging overlap relationships and truncation at frame edges. This gives the model consistent examples of partly hidden objects, which are common in driving, retail shelf and dense urban scenes.

For video, annotators draw polygons on keyframes and interpolate the shapes between them. Reviewers then check every interpolated frame and correct any drift. We place extra keyframes where an object changes shape, changes scale or is occluded, because interpolation breaks down in those places. The result is stable boundaries for tracking, action recognition and driving perception models.

Polygon annotations are delivered in COCO JSON, GeoJSON, LabelMe JSON, CSV, or client-defined schemas. Outputs include vertex coordinates, class labels, instance IDs and metadata, structured to load into PyTorch, TensorFlow and custom training pipelines.

We run large and ongoing projects in structured batches with a dedicated team and scheduled QA reviews. Work stays with specialist polygon annotators so boundary tracing stays consistent. When classes change, we version the taxonomy and re-annotate the affected batches, so quality holds over a long project.

Our workflows are ISO 27001-aligned, HIPAA-aligned, and GDPR-aligned, which matters for medical images, biometric data and other sensitive imagery. All annotators sign NDAs before accessing any project, and roles are permission-scoped. More detail is on our security practices page.

Polygon annotation cost is driven by image volume, object complexity, vertex density per image, occlusion frequency and review depth. Common models include per-image, per-object, hourly, or monthly retainer structures. Costs can be 40–60% below US/UK rates, depending on volume and complexity. For a detailed cost breakdown, see our data labeling pricing guide. You can also request a tailored quote based on your dataset type and volume.

Yes. We are platform-agnostic. Our annotators work within your internal tooling or any preferred third-party annotation platform, including CVAT, Labelbox, SuperAnnotate, V7 Darwin, and others. We adapt to your stack and workflow rather than requiring a platform switch.

We use in-house annotators rather than crowd-sourcing, and every project begins with a free pilot batch so you can check quality before committing. Our target of up to 99.8% accuracy, 3-tier QA process and ISO 27001-aligned workflows are covered on this page. For a broader vendor comparison, see our guide to top data annotation companies.

Yes. Polygon annotation for medical imaging is one of our most specialized use cases. We label organ boundaries, tumor contours, tissue regions, and anatomical structures across radiology, pathology and surgical imaging datasets, with clinician-guided reviewer validation and DICOM-compatible output. For geospatial projects, we deliver precise polygon masks for land parcels, buildings, roads and vegetation from satellite and aerial imagery. See the full scope on our medical annotation services.

Choose polygon annotation when your AI model needs to understand the precise boundary, shape and spatial structure of objects, not just their approximate location. Bounding box annotation is faster and lower-cost, but introduces significant background noise and is less suited to irregular, curved, or occluded objects. Polygon labeling matters most for autonomous driving (vehicle silhouettes, pedestrian outlines), medical imaging (organ and tumor boundaries), agriculture (canopy and crop contours), and any application where boundary accuracy directly impacts model performance.

Share your project brief (image type, object classes, volume and output format) through the contact form on this page. We'll assess complexity, agree annotation guidelines with you, and deliver a free pilot batch (typically 50–100 images) so you can evaluate polygon mask quality before committing. Pilot starts within 48 hours after the NDA is signed, and pilot delivery takes 24–48 hours, where applicable. Our team handles taxonomy setup, annotator training, QA configuration, and delivery.