Participants of this challenge should submit 14 PNG images, one for each of the test images, with value 0 for background pixels, and a unique positive integer for pixels corresponding to each segmented nucleus, similar to the label data provided for training images.

Aggregated Jaccard Index (AJI) will be used to compute the nuclei segmentation accuracy. The details of AJI are provided in Algorithm 1 of the paper provided in the Training Data section of the data page. 

  • Mean Aggregated Jaccard index over 14 test images will be computed to rank the participants
  • Submissions with missing results on any of the test images will not be ranked
  • Only fully-automated methods, that is the methods that require no manual intervention during testing, will be ranked 

The code to compute Aggregated Jaccard Index (AJI) is available here.