Ilastik not-responding

Hello! I am learning to use ilastik to perform the annotation of cell’s nucleuos but I am having some problems because ilastik freezes when I am runing the live update to get the segmentation of all the images, I am working with a image stak of 70 frames, the file is .tif format with a size of 309MB, also when I want to save the current project ilastik is not responding.

In this case, I am labeling frames 0, 17, 35, 51 and 69 to have a better segmentation process because the segmentation works better this way. Also, I have notice that the program stops working when the processing bar reaches 80%. Here I send and example of the images I am trying to label. Thank you in advance

Hi @xgalindo,

could you please confirm, that I am understanding correctly, what you want to achieve?

As far as I read it, you have a .tif time series (I assume it is a single file) with which you do pixel classification.
You do brush stroke annotations in a couple of frames, then you switch to live update. Here is where ilastik then is not responding, is that correct?

In any case, ilastik is not particularly good at reading tiffs - so you might want to convert your image to hdf5 using the ilastik/fiji import/export plugin.

Also I would disable live update before jumping between frames. ilastik will predict every frame when you move the slider. Just re-enable it once you’re in the frame you want to look at. Or just swich one frame forward/backward.

You were mentioning an example of the images you are labeling but embedding it here didn’t work, I think.

Thank you for your answer.

Indeed, I am using a single .tif time series file to do the pixel classification. My file is 71 frames, I brush stroke annotations in frame 0, 17, 52 and 70. I notice that live update makes everything slower so as you said I disable it before jumping between frames.

I am using .tif file because when I tried to use the ilastik plugin in fiji to convert my file from .tif to hdf5 ilastik did not recognize the new file when i tried to used it as input.

I will try to upload one of my files but it seems that is not working either

Oh, the fiji plugin should really work. It would be great to get hold of your image so that we can see where it goes wrong…

Thank you for your answer, Can I send you my data by mail? I have tried to uploaded it here several times but apparently I cannot do it here

Hey @xgalindo,

it would be great if you can upload it somewhere (dropbox or google drive) and send a link to team@ilastik.org .

Cheers
:slight_smile:

I already send you the email with the data, Thank you very much

Heyo @xgalindo,

I’ve sent you a link to one of your images that I converted to h5 via email. I could open it without issues in ilastik. So maybe please update the ilastik fiji import/export plugin in fiji.

For the problem at hand, there is always try to downsample (if the biological question you want to answer allows for that. But maybe you need this resolution). Is your resolution isotropic (so same size in all dimensions?).

Furthermore, when annotating I would maximize one of the views (since data is 3D you usually see 3 orthogonal ones).

And what you can also try, when you are in the training applet there is a setting in the menu: View -> set tile width. Per default this value is 512. For 3D data a value of 256 might work better. If you set it once, ilastik remembers it (for all other projects you use ilastik for as well).

Following up on an email here.

So I gave your image a go (the one that I converted to h5) and did not find that it would go into the non-responding part for me. Did you convert the rest of your images as well? Are you using multiple images in the project?

3D data is always heavy on the computer. There are a few things to keep in mind to make ilastik a bit snappier. I found a post here on the forum where I gave a quick summary: Which upgrade for my laptop would be most beneficial for working with "Ilastik"?
In addition to the above, you could also experiment a bit with the features you are using. E.g. try using all in 2D only.

And yes, in the one image I’ve seen, the segmentation looks challenging due to uneven contrast in your image and overlapping/touching cells.

Dear Domink

Thank you for your answer as always. Yes, indeed now I am using .h5 images that I converted using the ilastik fiji plugin. I am not using a laptop, instead, I use a Dell CPU with 64-bit Windows 10 OS, 128 Gb of ram 2 processors of 2.19 GHz and 1T hard drive.
I have not try to use the 2D features, I will do it today and I will let you know how it goes. Thank you very much for all your help.

Have a nice week