StarDist review
A focused, open-source toolkit for training and running 2D and 3D instance segmentation models.
Reviewed by iTechGuides Editors · Editorial team · Updated Oct 2026
StarDist is an open-source Python package for detecting and segmenting star-convex objects in 2D images and 3D volumes. It is aimed at image-analysis and research workflows, particularly microscopy researchers working on individual cells or nuclei. Rather than labeling whole regions semantically, it predicts object probabilities and boundary distances, then uses candidate polygons or polyhedra to produce separate instance labels. It runs on Windows, macOS, and Linux, and is available for self-hosted use.
Its strongest fit is a workflow that needs object-level labels and room to train models. Users can apply pretrained 2D models for fluorescence and H&E nuclei, or train with paired images and instance-labeled ground truth. Multi-class prediction is supported, as are evaluation metrics including precision, recall, F1, and panoptic quality. These capabilities suit research teams that need to assess instance-level results or adapt a model to their own labeled images. StarDist is not a semantic segmentation tool, so workflows centered on labeling broad pixel regions should look elsewhere.
The package connects with Fiji/ImageJ, napari, QuPath, Icy, and KNIME, which can make it a fit for teams already working in those image-analysis environments. The project is open source and free to use, with self-hosted deployment and community support. The Fiji/ImageJ plugin is documented for 2D or 2D+time images; that scope should not be treated as general video segmentation. Choose StarDist for specialized 2D/3D instance segmentation and model training, especially for nuclei work. Consider another option if you need semantic labels, managed deployment, or a different support model.
StarDist pros and cons
- Where it wins
- Segments instances in 2D images and 3D volumes
- Includes pretrained 2D models for fluorescence and H&E nuclei
- Integrates with Fiji/ImageJ, napari, QuPath, Icy, and KNIME
- Where it doesn't
- Specializes in instance segmentation, not semantic segmentation
- Training requires images paired with instance-labeled ground truth
- Self-hosted workflow with community support
StarDist fact sheet, pricing and score →
Advertiser disclosure: iTechGuides is reader-supported. Vendors can pay for top positions in our rankings and for a place on other products' pages, and we may earn a commission when you click some links. How we rank.
Last updated · How we research and update