Creating a new project

ilastik can work with separate 2D images and with stacks of 2D images, representing a cube of 3D data. Besides, it can handle the case if an image which has several channels and these channels are represented by different files. To start a new project and load the data, click the "New" button on the Project Tab:

You should now see a dialog which allows to set different project properties and load data files:

If you wish, set the project title, author(labeler) and description in the corresponding form entries. Different buttons below correspond to loading 2D and 3D data. Let's start with 2D.

Loading 2D images

The following two buttons allow to load 2D images into ilastik projects:

The "Add" button leads directly to the system Open Files Dialog. Simply choose the files you like, holding the "Ctrl" key as usual if you want to select multiple files. The "Load File" button allows to load an image which consists of several files, such as a multichannel image. This case is treated in the "Loading multichannel data" section.

Loading 3d image stacks

Start by clicking the "Load Stack" button. You should see a dialog like this:

Select the folder in which the images of the stack are stored. Note, that wildcards are supported for this input, so you can give something like "My data folder/My files from yesterday *.png". In the example below I selected all the files that have "z" in their title:

For quick feedback, ilastik looks at the first image of the stack and loads its size into the form. The number of selected images is taken as the Z-range of the data. If you want to select a subvolume of the data, simly change the "From" and "To" values of the volume coordinates.

Now let's take a look at the options.
  • The "Load MultiChannel data" option is treated in the "Loading multichannel data" section.
  • The "Invert Colors" option inverts the colors of the images in a stack. While it brings no advantages for the classification task as such, it makes some microscope images easier on the human eye and thus simplifies labeling. For example, compare the following 2 images:

  • The "Convert to Grayscale" option converts the images of the stack to 8-bit grayscale images.
  • The "Normalize Data" option stretches the range of the intensity values of the image between its maximum and minimum instead of between 0 and 255. This operation makes the low-intensity signal look brighter

  • The "Downsample" option downsamples the volume to the size specified by user in the widgets, which appear after you check this option. The downsampling is performed by spline interpolation of gaussian-smoothed images.

  • Finally, the "Save to Destination File" option allows you to save the volume into an .h5 file. This option is very useful, if you perform some operations on the data, such as normalization or downsampling and don't want to repeat it every time you load the stack. Besides, it allows you to keep your whole stack in one file.
  • The "Preview files" button shows a list of files which will be loaded into the stack. You might want to click it if you used wildcards to only select some files from a folder and want to be sure that the selection has worked correctly.

Loading multichannel data

2D images

After clicking the "Load File" button, you should see a dialog like this:

Suppose, you have an image with 3 channels. Select the path to the first one as usual and then click on the "Append more spectral channels" button:

A new line will appear, where you can enter the path to the file, containing the second channel.

Then simply continue clicking the "Append more spectral channels" button until all your channels have been added. After the paths to all channels have been provided, click "Load".

3D stacks

If you want to load a stack of multichannel images, check the corresponding checkbox:

You should see a line, where you can type the identifier of the channel. Usually, if a series of images is taken in multiple channels, the files which belong to the same channel have this channel as part of their name. For example, all the files for the first channel can end with "GFP" or "channel001". To add more channels, simply press the "Add channel identifier" button. Here is an example of loading a 3-channel stack:

To check, that the right files have been picked up, click the "Preview files" button:

Viewing multichannel data

By default, ilastik will blend the 3 channels into one RGB image:

If you want to examine the channels individually, uncheck the corresponding option in the dialog, which pops up after you click the "Options" button:

Once this option is unchecked you will see a channel-selecting widget in the bottom right corner:

The view will switch to individual channels if you change channels in this widget.

Project Tab options

  • The "New" button allows to create a new ilastik project. The dialog, which appears after you click this button, is described in "Creating a new project section" .
  • The "Open" button lets you load a previously saved project
  • The "Save" button saves the current project into a file with .ilp extension
  • The "Save as" button allows to save the current project under a different name
  • The "Edit" button lets you change all project properties: title, author, description and even the data. Use the "Remove" button to remove the images you don't want anymore and "Add" and "Load" buttons to load new files.

  • The "Options" button shows several other options which relate to the displaying and navigation of images. Let's go through them one by one.
    • "Train and predict in Interactive Mode" checkbox describes ilastik behavior in interactive prediction mode of the Classification Tab. If this checkbox is checked, the classifier will take in new labels every x seconds, where x is specified by the user with default 300ms.
    • "Normalize data in each slice view" checkbox allows to normalize the intensity range to the minimum and maximum of each slice separately, instead of the overall minimum and maximum of the data.
    • The "Interpret 3D channels as RGB" checkbox is described in "Viewing multichannel data" section.
    • The "Show border margin indicator" checkbox also corresponds to Classification Tab connected behavior of ilastik. It allows to visualize the border region, where the feature computation is based on mirrored data values and the classification is not as reliable.
    • The last checkbox, "Speed up painting of slice views by tolerating flickering" is self-descripting. Try ilastik with and without this option and see if the flickering is too much for you.