This is maybe a silly question. In the sample submission file, x, and y are corner of the bounding box, right? so how about the z? is that the center slice of the bounding box or the corner (and the depth is the length in the z dimension)? For the training data, the z seems to indicate the center of the slice for the bounding box.Posted by: quanchen @ Jan. 19, 2021, 9:39 p.m.
Z (first not center) to Depth should be the entire lesion range. In the training boxes file "Slice" was the center. You won't know what the center was, but it will be used to calculate +\-25% of the volume slices to identify the true lesion range. That will be compared to your range of [Z -> Depth].
Z: integer - the first bounding box slice number in 0-based indexing (for the first slice of the image z=0)
Depth: integer - predicted bounding box slice span (size along the depth axis)
When I said training data I was referring to the boxes file. To practice you can take the slice # in that file and add\subtract 25% of volume slices and that should be your ideal Z to Depth range for classifications. For obvious reasons, we can't share that information in the validation and test phases.Posted by: challenge-organizer @ Jan. 19, 2021, 10:19 p.m.