Image Segmentation Tool

Create segmentation labels using a powerful labeling interface. Easy to set up and integrate into your workflow.

Try for free No credit card required
Trusted by ML teams from...

Smart labeling tools

Label objects and regions with unseen speed and accuracy, assisted by our Superpixel and Autosegment technology.

Learn more

Semantic, instance, or panoptic labels

Export your labels as instance, semantic, or panoptic segmentation labels. Multiple common export formats supported.

Learn more
Instance vs. semantic segmentation
Model predictions vs. corrected labels

Model-assisted labeling

Label training data faster by leveraging your model. Upload model predictions for new data and correct the predictions using smart labeling tools.

Learn more

Why ML teams choose

  from segments import SegmentsClient

  client = SegmentsClient("api_key")

  dataset = client.add_dataset(

  sample = client.add_sample(
TensorFlow, AWS, Hugging Face, and PyTorch logo

Easy to set up and integrate

Integrate data labeling into your existing ML pipelines and workflows using our simple yet powerful Python SDK.

Connect your cloud provider (AWS, Google Cloud, Azure)

Export to popular ML frameworks (PyTorch, TensorFlow, 🤗 Hugging Face)

Manage your workforce

Collaborate with your team or onboard an external workforce. Our management tools make it easy to build and review large datasets together.

Give collaborators access to specific datasets and choose their permissions

Onboard an external workforce from our labeling service partners

Set up a reviewing step and communicate with labelers via issues

Track labeling performance through metrics and custom dashboards.

Model predictions vs. corrected labels
Model predictions vs. corrected labels

Set up advanced workflows

Implement active learning by using your model predictions to find failure cases and to speed up labeling.


Label images, point clouds, and text on


Train a model and keep it on your side. Only upload model predictions


Compare predictions with ground-truth labels and use our search syntax to find failure cases


Use model predictions as prelabels to speed up labeling

Start building better datasets today

Try free for 14 days. No credit card required.

Start free trial