Binaries
Choose from the download links below, and make sure your operating system satisfies the minimum requirements.
For help with installation, see the installation instructions.
Version 1.4.0.post1 (10. November 2023)
Release Highlights:
- Neural Network Workflow: run pre-trained neural networks from Bioimage.io Model Zoo in ilastik
- Networks can run on CPU, GPU (choose the GPU build for Windows, or Linux), or on a remote server
- No Neural Network training, yet - we’re working on it.
- New startup screen!
- Major UI and performance improvements for the Multicut Workflow
- Stability/Usability improvements in the Carving Workflow
- By popular demand: redo/undo for brush-strokes
- Many bugfixes throughout
ilastik 1.4.0 is distributed in two versions for Windows and Linux.
In addition to our regular build, you can download GPU-enabled builds.
With these you can
run pre-trained neural network models from the
bioimage.io Model Zoo faster, if you have a compatible nvidia GPU on your system.
All other workflows, besides the
Neural Network Workflow (local) will work on the CPU, like before.
If you are interested in running the ilastik Neural Network Workflow on a
remote GPU, e.g. if you have a powerful server in your facility, you have to check for additional installation instructions for the server part in the
tiktorch repository.
After you install the server part, you can use your regular ilastik installation as a client with the
Neural Network Workflow (remote), ilastik will take care of the communication with the server.
Commercial solver support with gurobi 951 for tracking with learning.
Note that on OSX there is no 3D preview in Carving, all other functionality should work as expected.
We recommend OSX users to download the current
beta version.
If you have an Apple Silicon (M1/M2/M3), then please select the "Apple Silicon (arm64)" build.
Regular builds
GPU-enabled builds
For faster neural network prediction.
ilastik GPU-enabled builds are distributed with NVidia's CUDA toolkit. By downloading and using these builds, you accept the terms and conditions of the
CUDA End User License Agreement (EULA).
Beta Version 1.4.1b21 (8. October 2024)
Release Highlights:
- Added Trainable Domain Adaptation Workflow
- Native build for Apple Silicon (M1, M2, M3)
- Tiff reading improvements
- More responsive user interface in Multicut
- Added spherical texture features, contributed by Oane Gros (preprint)
Notes:
- Commercial solver support with gurobi 951 for tracking with learning.
- This beta version is currently distributed without tensorflow
- Experimental native build for Apple Silicon (M1, M2, M3) is available, which means better performance on these machines overall.
Notes for the silicon version:
- Currently only 2D networks are supported in the Neural Network and Trainable Domain Adaptation Workflows.
These, however, run accelerated leveraging Metal Performance Shaders (mps).
Regular builds
GPU-enabled builds
For faster neural network prediction.
ilastik GPU-enabled builds are distributed with NVidia's CUDA toolkit. By downloading and using these builds, you accept the terms and conditions of the
CUDA End User License Agreement (EULA).
Previous stable versions
macOS Big Sur and Monterey users please download ilastik 1.4.
Dear macOS users, you are running macOS 11 (Big Sur), or newer please download
the latest ilastik version (1.4.0 or newer).
Older versions don't run anymore on OSX.
Version 1.3.3
Includes various improvements and bug fixes.
A few highlights:
- 1.3.3post3, projects are opened readonly in headless mode per default
- Support for n5 volumes (https://github.com/saalfeldlab/n5)
- Support for masks in Pixel Classification (an additional dataset can be supplied that disables computation of probabilities wherever it is 0 -> speeds up computation in cases with a lot of background
- Atlas Mask overlays for Object classification (an additional dataset can be supplied. Pixel values of this additional mask are included in result tables but not used in classification.)
- counting fixes (saving boxes, import/export of boxes, interaction improvements)
- autocontext now takes arbitrary data types (was previously limited to uint8), but mind the memory footprint! We still recommend to use uint8.
- carving improvements, more efficient memory utilization
- applet bar improvement, redesign
- a lot of internal backend cleanups/improvements
- a lot of internal GUI cleanups/improvements
- updated dependency stack
For more details, see the
Release Notes.
Commercial solver support for cplex version 1290 and gurobi 811 for tracking with learning.
Version 1.3.2
Please see our
Release Notes for a summary of changes.
Version 1.3.0
We have updated all our dependencies, and switched from Python 2 to Python 3.
This could mean that projects created with older ilastik versions do not load any more and vice versa.
We advise you to make copies of important project files before you try to use them with this new version of ilastik!
For more details, see the
Release Notes.
Version 1.2.2 (based on Python 2)