Hi, I was not able to find a reference for Multiple Instance Dice score and Multiple Instance Normalized Surface Distance. Can you please give a description or reference? The reference you have given on wiki page does not contain the information or have I misunderstood? Since we are not doing classification, and same instrument instance can take different label between different ground truths (according to its position I believe?) , I am a bit confused how the evaluation will be done. There is a similar problem also in Multiple Instance Detection evaluation but I understand the evaluation in that case will be based on IOU, so that the location will be considered rather than the label of instances.

Created by kkirtac
Dear kkirtac, we are referring to the "Assessment Criteria" section. Both metrics are implemented and a python script is provided there which will also be used for the ROBUST-MIS challenges. Kind regards, the ROBUST-MIS organizers

Multiple Instance Segmentation evaluation page is loading…