Here, a brief summary of answers for some questions during the webinar. (feel free to add missing)

  • why are foreground masks provided mostly for technical reasons, to reduce computation time so that participants have it easier to have <1h compute time. especially for l1 this should save a lot of computation time, as the rois are quite small
  • tumor-associated stroma and normal stroma they are visually very similar, the difference may be sometimes only the presence of tumor nearby. To find examples of healthy stroma you could take the tissue outside of the annotated bulk
  • maximal size of tissue in the slides: ~470mm². (l1 masks cover only the rois and are therefore <5mm²)
  • froc - score computation we defined the score-parameters empirically based on (early) results from the baseline. see "Evaluation" for details
  • is it allowed to use something beside segmentation/detection for l2? yes (if in doubt, consult the Rules section :-). Since the same algorithm will be applied both to l1 and l2 you might still want to have your algorithm segment and detect to have a chance at the l1 prize.