Can Pre-trained model be improved?

Dear Oliver,
Looking at the images there are very promising it is incredible how can separate overlapping cells. I test 20 images from which I already know how much cells there are from manual counting and I compare with the prediction made from stardist , the results are not so good. Can you help me suggesting me some specific changes that I can make in terms of parameters or similar and try to see if this can improve the pre-trained model to make it perfect for my data?

Thank you very much for your help and for what you are doing.


Dear @Mariya_Timotey_Mitev,

This would mean you must re-train a model to detect the cells. To do that you should annotate about 10 images, perhaps @superresolusian has a protocol for annotating in Fiji?