DOCSEARCHbyAlgolia
Submit Docs

ilastik Documentation

website·Indexed 35 days ago

https://www.ilastik.org/documentation/index.html

Documentation for ilastik, a software suite for image analysis featuring workflows for pixel classification, voxel segmentation with active learning, animal tracking, and interactive counting. Includes guides on headless mode operation, batch processing, performance optimization, and integration with Fiji via a dedicated plugin. Covers versions up to 1.4.2, including specialized tools like Trainable Domain Adaptation and IIBoost Synapse Detection.

Tokens
30.4K
Snippets
42
Records
225
Agent score
72%

What's inside ilastik

  1. Overview of ilastik Object Classification workflows

    The Object Classification workflow classifies full objects (sets of pixels belonging to the same instance) based on object-level features and user annotations. It requires raw image data and a second input that identifies object boundaries. There are three workflow variants based on the second input type:

    1. Object Classification [Inputs: Raw Data, Pixel Prediction Map]: Used when you have a probability map (e.g., exported from the Pixel Classification workflow) where each pixel value represents the probability of belonging to an object.
    2. Object Classification [Inputs: Raw Data, Segmentation]: Used for binary segmentations (e.g., created via thresholding in Fiji). Background pixels must have a value of 0.
    3. Object Classification [Inputs: Raw Data, Label Image]: Used for inputs where each object already has a unique identity label (e.g., outputs from Stardist or Cellpose). Background pixels must have a value of 0.

  2. Overview of the Carving workflow

    The Carving module uses a seeded watershed algorithm for interactive object extraction. Unlike Pixel Classification, which relies on internal appearance (brightness, color, texture), Carving relies on discernible object boundaries. It is specifically designed for objects that do not differ from their surroundings internally but are separated by clear boundaries (e.g., cell membranes in electron microscopy images).

  3. Overview of the Animal Tracking Workflow

    The Animal Tracking Workflow in !ilastik is designed to track lab animals (e.g., flies, larvae, mice, or zebrafish) in 2D+t or 3D+t videos. The workflow automatically assigns IDs to objects based on sparse labels and user-defined parameters. It is capable of handling object appearances, disappearances, over-segmentation, and mergers (clusters of objects). Project settings, including labels and parameters, are stored in a .ilp project file for reuse in batch processing pipelines.

  4. Overview of the ilastik Fiji plugin

    The ilastik Fiji plugin integrates ilastik with Fiji, primarily providing the ability to import and export data in HDF5 format and run trained ilastik workflows directly within Fiji. HDF5 is recommended for large datasets, particularly multipage TIFFs, to enable efficient block-wise processing.

    Supported workflows include:

    • Pixel Classification
    • Autocontext
    • Object Classification
    • Multicut (Boundary-based Segmentation)
    • Tracking

    All plugin features can be automated using Fiji macros.

  5. Overview of ilastik Pixel Classification

    The Pixel Classification workflow performs semantic segmentation by assigning labels to pixels based on selected features and user annotations. It uses a Random Forest classifier to generate probability maps for each class. Note that this workflow provides semantic segmentation (class-based) rather than instance segmentation (individual object-based). To convert probability maps into individual objects, users can use thresholding and connected component analysis (available in the Object Classification Workflow) or external tools like Fiji for watershed and agglomeration algorithms.

  6. Overview of Boundary-based segmentation with Multicut

    The Multicut workflow segments images by using a boundary probability map to break the image into superpixels and then merging them to recover segments limited by closed surfaces. This approach prevents 'dangling edges' and is particularly useful for biological image analysis, such as connectomics, membrane staining, or phase contrast images.

  7. Overview of ilastik image analysis capabilities

    ilastik is a modular software framework designed for interactive image classification, segmentation, and analysis. It provides workflows for supervised pixel- and object-level classification, semi-automated object tracking, segmentation, and object counting. To optimize performance, most analysis operations are performed lazily, allowing users to interactively process data subvolumes before running a complete volume analysis in offline batch mode.

  8. Overview of ilastik Object Classification Features

    ilastik uses object features (numerical descriptions of objects) to differentiate between different types of object classes during classification. By default, ilastik includes three feature plugins: 'Standard Object Features', 'Skeleton Features' (available for 2D only), and 'Convex Hull Features'.

  9. Overview of Autocontext cascaded pixel classification

    Autocontext is a multi-stage extension of the Pixel Classification workflow. It improves segmentation by running classification in multiple stages, where the output of each stage is added as new channels to the raw data and used as input for the next stage. The algorithm not only uses previous results during prediction but also computes features based on those results. This is particularly effective for images with multiple distinct classes (e.g., separating membranes from cytoplasm and organelles in EM images).

  10. Overview of ilastik image analysis workflows

    ilastik provides several specialized workflows for image segmentation and analysis:

    • Pixel Classification: Divides all pixels into user-defined classes based on sparse example annotations and pixel-level features.
    • Autocontext: Enhances Pixel Classification via cascaded classification, where the results of a first classification round serve as features for a second round.
    • Object Classification: Assigns objects (extracted from a segmentation mask) to user-defined classes using object-level features and example assignments.
    • Tracking: Tracks objects across the time axis for 2D and 3D data. On Windows, the automatic tracking workflow requires the external CPLEX library for optimization.
    • Animal Tracking: Specialized tracking for lab animals (e.g., flies, mice, larvae, zebrafish) in 2D+t or 3D+t videos.
    • Density Counting: Counts objects in 2D images without requiring prior segmentation, effective for crowded images with overlaps.
    • Carving: Semi-automatic interactive segmentation of 2D and 3D data using sparse foreground/background seeds.
    • Boundary-based Segmentation with Multicut: Segments objects based on boundary information (ideal for electron microscopy or membrane staining) by solving an optimization problem to create closed-surface objects.

  11. Overview of Interactive Density Counting

    The Interactive Density Counting workflow is designed to count objects in crowded scenes (e.g., cells in microscopy images) where objects may overlap. Unlike standard segmentation, which often underestimates counts due to under-segmentation in dense areas, this workflow estimates the object density directly. It is most effective for counting many blob-like overlapping objects with similar appearance, size, intensity, and texture.

  12. Overview of ilastik Workflows

    ilastik provides several specialized workflows for image analysis and segmentation:

    • Pixel Classification: Divides all pixels into user-defined classes using sparse example annotations and pixel-level features.
    • Autocontext: Enhances Pixel Classification via cascaded classification; the results of a first round of pixel classification serve as features for a second round.
    • Object Classification: Assigns objects (extracted from a segmentation mask) to user-defined classes based on object-level features and example assignments.
    • Tracking: Tracks objects across the time axis for 2D and 3D data. On Windows, this requires the external CPLEX library for optimization.
    • Animal Tracking: Specialized tracking for lab animals (e.g., flies, mice, larvae, zebrafish) in 2D+t or 3D+t videos.
    • Density Counting: Counts objects in 2D images without requiring prior segmentation, effective even in crowded images with overlaps.
    • Carving: Semi-automatic interactive segmentation of 2D and 3D data using sparse foreground/background seeds.
    • Boundary-based Segmentation with Multicut: Segments objects based on boundary information (ideal for electron microscopy or membrane staining) by solving an optimization problem to create closed-surface objects.