Reading your code I can't see how you're using multitask learning. It isn't well explained in the paper in my humble opinion. Can you elaborate on this? Where does the cell count label 'c' come from? Is there a dedicated output layer to generate this label? Surely you don't do mean shift clustering at every training iteration?
Reading your code I can't see how you're using multitask learning. It isn't well explained in the paper in my humble opinion. Can you elaborate on this? Where does the cell count label 'c' come from? Is there a dedicated output layer to generate this label? Surely you don't do mean shift clustering at every training iteration?