Adding seeds to ilp (hdf) file automatically

Hello! I really like Ilastik to run quick prototypes. But now I have a bigger project but not big enough to warrant its own software development. So I have hundreds (maybe thousands) of average small images (258x258) and their corresponding seeds. It is unfeasible to load them manually one by one.

I tried to check the ilp file inside (I understand it is a HDF file?) to see if I could add the images and labels for training in an automatic way but I couldn’t open the ilp file. Do you have any suggestion on how I could do this?

Hi @lesolorzanov,

just a quick question, the annotations you have for the images, are those dense (most pixels labeled) or sparse (only a few pixels labeled).

it is a hdf5 file, indeed. The actual format of the data is a bit intricate and reflects the implementation of the annotation cache in ilastik.

Are those images all the same size? Another approach would be to convert those to a time series (as well as the annotations) and then import the labels via the functionality that is available on right click on the label layer in the training applet.

Ohh that sounds interesting. They are not dense. there is quite a lot of background. (I attach a screenshot). I create a project with the correct amount of labels and add a few examples and then I was looking, naively, for a level or variable containing such arrays, silly me.

convert those to a time series

How would I go about that? do you know? A multilayered tiff? a video? a specific format? hdf? This sounds like a good temporary solution.

are you using Fiji? Or do you want to code something to do the conversion?

In fiji you’d import your files via File › Import › Image Sequence. Then the easiest would be to export it using the fiji/ilastik import/export plugin.
You would do this once for the raw data, and once for the annotations (making sure the order in time will be the same).

1 Like

Perfect, I created the images with the Ilastik plugin for imageJ and imported them without problem. Thank you very much!

Great to hear @lesolorzanov :slight_smile: