ilastik

ilastik does interactive machine learning for image segmentation and classification: you paint a few example strokes on the structures you care about, it trains a classifier on the fly and shows you the result immediately, and you correct where it is wrong. No coding, no training set assembled in advance.

What it does

The workflow is genuinely interactive — the classifier updates as you annotate, so you spend your effort on the cases it gets wrong rather than labelling exhaustively. Beyond pixel classification it covers object classification, counting, tracking and boundary-based segmentation, and trained projects can be applied headlessly in batch across a whole dataset.

Strengths

• Free and open source, with no programming required.
• Interactive training converges in minutes on many problems.
• Batch application of trained classifiers via headless mode.
• Effective on noisy or low-contrast images where thresholding fails.
• Works on 2D, 3D and time-series data.

Limitations

Classifiers trained on one imaging session often generalise poorly to another with different acquisition settings — retraining is routine rather than exceptional. Classical features mean it is now outperformed by deep-learning segmentation on hard problems. Memory use on large 3D volumes is heavy, and the annotation you provide determines the result entirely.