Ilastik structured tracking fail to detect cell division

When using structured learning-based tracking model, I find it cannot attribute child cells into the same lineage, even though parent and child cells have been assigned in training applet. Does anyone know if it is correct?

(In training applet, two child cells are labeled in red and green; tracking parameter are shown)


Expected: daughter cell in 2 new tracks belonging to the same lineage of mother track.

Hi @Jeff-Gui,

the structured tracking with learning workflow has this applet in order for users to annotate example tracks on which the parameters for solving the tracking problem are learned. The “calculate tracking weights” Button is greyed out in your screenshot, so do I assume correctly, that you didn’t train the parameters? Did you change them by hand? In order to train these parameters (and with that take into account the annotations you have added in the “Training” applet/workflow step). See here for a description of the learning part in tracking.
Furthermore, in order to use this, you’ll have to install a commercial solver. We currently recommend Gurobi, which is free for academic use. We have some information on how to obtain such a solver here in our docs.

Does this help you?

Hi, that really helps. I’m applying for the academic license for some solver. Without a solver installed, I could not press “calculate tracking weights” button therefore edited these parameters.

I have another question regarding common tracking model. It seems that ilastik sometimes transfer cell identify to a far away object when the cell is moved outside of the frame. For example, the dark green cell is dividing while the light green cell is disappearing in the picture. The newly born cell is assigned an identity of the disappearing one, not daughter cell. In this case, how can I avoid transition by changing parameters? Thanks.