Roadmap

This roadmap is supposed to give an overview on the current state of ilastik and things to expect in the near future. It also stores ideas for new developments, which are too large to fit in a single github issue.

In code, but not yet (fully) enabled or documented

• Possibility to switch classifiers in pixel classification - in debug mode already present
• Lazy connected components from python side

Coming soon

• Interactive feature selection
• Structured learning for tracking
• Multicut workflow for globally optimal image segmentation

What else we'd like to have

• Read special hdf5 formats, e.g. from Fiji, Matlab
• Import 5D data from lists of pngs, e.g. as list of files named vol-T##-Z###-C##.png
• Error correction in carving: assign supervoxels to objects by clicking
• Documentation as protocols, extending to other packages. E.g. how to segment cells in ilastik and then do analysis in FiJi.
• Variable importance for classification (as computed by random forest) - see pull request
• Navigate to labels
• Simpler object feature selection dialog with feature groupings
• Overview of prediction results as a histogram of classes over time or over the stack
• Multiscale views. More details here.
• 2d/3d object browser and mesh export for workflows other than carving. More details here.
• OpenGL blending in volumina. More details here.
• Plugins for pixel classification features
• Rethink the headless interface: JSON input
• "Fancy" 3D visualization (e.g. the translucent volume rendering from NeuTu)
• "Dummy" workflow for individual applet testing
• GUI for switching classifiers, with custom parameter dialogues for each classifier
• QT5 and python3 - which dependencies support it already?
• Show the same view in all applets
• Start object classification from connected components
• Clusterized version of ilastik