ilastik
Interactive machine-learning toolkit for pixel and object classification with sparse annotation.
Overview
ilastik trains classifiers from a few user-drawn brush strokes and applies them across full datasets. It's useful when thresholding or classical segmentation isn't separable enough but you don't have (or want) a labelled training set for deep learning.
Key features
- Pixel classification and object classification workflows
- Autocontext, carving, and tracking workflows for harder cases
- Headless/batch mode via CLI
- Works on 2D, 3D, and time-series data
Installation
Download the installer for your OS from ilastik.org/download. ilastik is distributed as a self-contained application; no Python environment needed.
Integration with other tools
- Fiji: apply a trained ilastik project to many images via the ilastik ImageJ plugin.
- TrackMate: use ilastik segmentations as input to tracking via the TrackMate-ilastik detector.
- CellProfiler: load ilastik predictions as input to a CellProfiler pipeline.
Official documentation
Learning resources
Written
- Pixel classification tutorial — the most common workflow, good starting point.
- Object classification tutorial
- Tracking workflow