The evaluation script (https://github.com/carpenterlab/2019_caicedo_cytometryA/blob/master/unet4nuclei/utils/evaluation.py) contains a step to calculate the number of objects:
# Count objects
true_objects = len(np.unique(ground_truth))
pred_objects = len(np.unique(prediction))
However, this will also include the background in the object count (it has the unique value 0).
To fix this change the lines to:
# Count objects
true_objects = len(np.unique(ground_truth)) - 1
pred_objects = len(np.unique(prediction)) - 1
Or is the background included here on purpose as it is removed from the analysis at a later step?