Written and maintained by CASRAI Editorial Board
Last updated
ilastik is a free, open-source toolkit for interactive machine-learning image analysis. You label a few example pixels or objects, the software learns from them, and it applies the result to segment, classify, track or count structures in your images, with no machine-learning expertise required. This page is for researchers and core-facility staff deciding whether ilastik fits an imaging project and how to document and cite it. Information below comes from the project’s own pages, read 2026-11-08.
What is ilastik?
The ilastik website describes it as “the interactive learning and segmentation toolkit”. It uses machine learning to segment, classify, track and count cells or other experimental data. Instead of writing an algorithm, you annotate examples directly in the interface and the software trains a classifier from them. Installers are provided for Linux, Mac and Windows. The home page announced version 1.4.2 on 28 May 2026.
Who runs ilastik and how is it licensed?
The ilastik team develops the software in Anna Kreshuk’s lab at the European Molecular Biology Laboratory (EMBL). Its About page says funding came partly from the DFG, the Human Brain Project and SFB 1129, and that this funding does not imply endorsement of the software. ilastik is distributed under the GNU General Public License, version 2 or any later version, with a special exception that allows extensions of ilastik that are not covered by the GPL.
Which ilastik workflows exist?
ilastik is organised into workflows, each a guided sequence of steps for one kind of task. The documentation overview lists:
- Pixel Classification
- Autocontext
- Object Classification
- Tracking and Animal Tracking
- Density Counting
- Carving
- Boundary-based Segmentation with Multicut
- Neural Network Classification and Trainable Domain Adaptation
The documentation marks Voxel Segmentation (beta) and IIBoost Synapse Detection as deprecated. As an editorial reading of the workflow names: Pixel Classification is the usual starting point when you need to separate structures from background; Object Classification works on objects that are already segmented; the Tracking workflows follow objects over time; and Density Counting addresses counting. Confirm the input requirements of a workflow in its documentation page before committing a dataset to it.
How do you use ilastik for pixel classification?
The Pixel Classification documentation walks through these stages, using the interface’s own terms:
- Raw Data. Load the images to analyse.
- Feature Selection. Use “Select features” to choose colour/intensity, edge and texture features at several scales. “Suggest Features” can propose a set.
- Training. Use “Add Label” to create a class for each kind of structure, then scribble on example pixels. An eraser corrects mistakes and the Label Explorer helps review labels.
- Live update. Switching it on trains a Random Forest classifier and overlays the predictions so you can add labels where the prediction is wrong.
- Segmentation. Enable the Segmentation checkbox to view a simple segmentation derived from the predictions.
- Batch processing and Exporting Output. Apply the trained project to further images and export probability maps or a simple segmentation. A headless mode lets the same project run on compute clusters.
How do you cite and document ilastik?
The ilastik About page gives this reference: Berg S, Kutra D, Kroeger T, Straehle CN, Kausler BX, Haubold C, Schiegg M, Ales J, Beier T, Rudy M, Eren K, Cervantes JI, Xu B, Beuttenmueller F, Wolny A, Zhang C, Koethe U, Hamprecht FA, Kreshuk A. “ilastik: interactive machine learning for (bio)image analysis.” Nature Methods 16(12):1226-1232, 2019. doi:10.1038/s41592-019-0582-9.
For reproducibility, an editorial suggestion: record the ilastik version, the workflow used, the selected features and the exported project file alongside the images, and deposit them with the dataset. If the term is new to you, see our explainer on what machine learning is.
Strengths and limits
- Strength: labels are drawn directly on images and predictions update interactively, so no coding is needed for the Pixel Classification workflow (ilastik documentation).
- Strength: batch processing and a headless mode support applying one trained project to many images (ilastik documentation).
- Limit: results depend on the labels you provide, so the labelled examples should cover the variation in your data. This is an editorial point rather than a statement from the project.
- Limit: several older workflows are deprecated, so check the documentation overview before choosing one.
What are alternatives to ilastik?
Other image-analysis tools overlap with parts of ilastik’s scope. CASRAI covers these in separate pages:
| Tool | CASRAI page |
|---|---|
| QuPath | QuPath guide |
| Fiji and ImageJ | Fiji vs ImageJ |
| Fiji colocalization workflow | Colocalization analysis in Fiji |
For labelling images as training data in general, see image annotation. For the imaging method that produces typical ilastik input, see confocal microscopy.
Frequently asked questions
Is ilastik free?
It is distributed under the GNU General Public License, version 2 or later, with an exception for extensions not covered by the GPL (ilastik About page, read 2026-11-08).
Which operating systems does ilastik support?
The project provides packages for Linux, Mac and Windows (ilastik home page, read 2026-11-08).
Do I need to know machine learning to use ilastik?
The project states that no machine-learning expertise is required; you train the classifier by annotating examples.
How do I cite ilastik?
Cite Berg et al., Nature Methods 16(12):1226-1232, 2019, doi:10.1038/s41592-019-0582-9, as listed on the About page.
Sources
- ilastik home page, read 2026-11-08
- ilastik documentation overview, read 2026-11-08
- ilastik Pixel Classification documentation, read 2026-11-08
- ilastik About page, read 2026-11-08
