How icarlosvega/ilastik works
Project type and purpose
• From the README and file tree, this is a complex, desktop‐based interactive image analysis toolkit (“ilastik”) that guides users through machine‐learning workflows (pixel classification, object tracking, etc.). It is not a web application but a full‐featured GUI application with headless/CLI support and a pluggable operator/dataflow engine (“lazyflow”).
File structure
• Top‐level modules:
• ilastik/ ← main application (shell, applets, workflows, utility, plugins)
• lazyflow/ ← core dataflow engine and operators
• bin/ ← command‐line scripts
• notebooks/ ← example notebooks using APIs
• dev/, conda‑recipe ← development environment and packaging
• tests/ ← automated tests
• docs/ ← Sphinx documentation for ilastik and lazyflow
• Patterns:
• Modular “applets” under ilastik/applets – each implements one workflow step (GUI + operator + serializer)
• “Workflows” under ilastik/workflows – orchestrations of applets into end‑to‑end pipelines
• Core dataflow graph under lazyflow, with operators for I/O, filters, ML, caching
• Plugin manager handling third‑party feature/format plugins (ilastik/plugins and plugins_default)
• Deployment/configuration: setup.py, pyproject.toml, conda recipes, CI configs, packaging scripts in bin/ and scripts/.
Extract architecture hints from the README
• ilastik.org workflows are interactive but also headless (bin/train_headless.py, shell/headless).
• Relies on Qt for GUI (drawer .ui files under applets and shell/gui/ui).
• ML dependencies: vigra, scikit-learn, PyTorch via tiktorch.
• I/O supports HDF5, TIFF, RESTful volumes, DVID.