I recently tried Ilastik to track bacteria moving and it looks very promising. However, I cannot get it to do the Object counting classification. I am using the Animal tracking [Inputs: raw dat, prediction maps]. I could create the probability files and it seems they load correctly (from what I see in Input Data and Threshold and Size Filter). When I go to Object Count Classification and try to use the brush to select objects and classify them, nothing happens. I don’t get an error either. At the moment I am using pretty big files (200 frames), but I had the same issue with a small test case (10 frames). There, it automatically resolved after few trials, but I cannot recall what I did exactly.
what version of ilastik are you using?
so, I assume the thresholding went fine? So your objects show up with a mask (where each shape has it’s own color) in the thresholding applet and all the shapes are white in the object count classifier?
do you see an error when you try to mark something, or is just nothing happening?
Thanks a lot for your quick answer!
I am using Ilastik 1.3.2 on a machine with Windows 10, 64 bit. nothing was happening, no error of sort.
It looked like the thresholding was fine (the probability file was created and I could further threshold, smooth etch) but even the objects were not showing up like a mask with different colours.
I found a work-around, but I am not sure of why it works: my particles are max 300 pixels big, given the 1:1 calibration that I set via Fiji pre-processing. If I put the highest possible value (something like 1000000) in the max size of particles, the particles appear and then I can run the count classifier. I wonder if I have some wierd metadata saved in my files that I am not aware of and that Ilastik takes up automatically.
we do not really use a lot of meta-data. Values are in pixel-sizes. Might it be possible, that you are using the wrong input channel? The input channel selector allows you to chose between the differen classes that you generated in pixel classification (so there should be at least two).
in the above image I had to select the blue channel in order to get segmentations for the small objects, that are then highlighted in multiple colors as a result of the thresholding.
Yes, you are right! I had the wrong input selected. Now it works fine. Thanks!