how to transform segmentation masks to instance masks

[fabiofumarola]

Right now we are using the function groups, n_groups = scipy.ndimage.label(mask) to detect the groups of instance from masks, but we need to:

1. compute the confidence per group and not per global mask
2. add min, 1 quartile, 2 quartile, mean, median, 3 and 4 quartile

show the segmentation mask as heatmap on the image and see if there are areas of confidence

3. maybe we can have different annotations with different confidence per mask