Export and import of predictions

Hello there,

In order to avoid using autocontext (slow), but take advantage of double annotation I exported the probabilities of a pixel classification (two labels) as unsigned 16-bit tiff (input image is in the same format). Then I made a new pixel classification project and tried to load the previously exported probability and the same input image as a stack over C (channel). But what I got was an error message “list assignment index is out of range”. Any help/tips would be greatly appreciated!


Hello @Jama,

First of all, great idea to gain some performance in autocontext!
I just tried (with the current latest version, as well as 1.3.3post3) the stacking it and got the same error - :confused: . I’ve opened an issue to keep track of this problem here: stack import fails · Issue #2409 · ilastik/ilastik · GitHub

Okay I checked a bit more and while the error message is not very helpful, it looks like the code currently is not expected to stack images that have dimensions that are not matching.

As a … workaround I’d propose stacking the images in imagej, and exporting the composite. (Or write a small python script).

Thank you! I was just trying to adapt the strategy you mentioned in the object classification part of your webinar : https://www.youtube.com/watch?v=_ValtSLeAr0&t=2040s