Based on the definition provided (see below), all I need to do is to have a predicted box as large as possible (while not changing the center position) so that the center of the ground truth box is within half of the diagonal of my predicted box. Maybe you have to modify the criteria 1 as within 100 pixels (which is similar to the third dimension criteria)?
"Definition of a true-positive detection:
A predicted box is going to be counted as a true positive if the distance in pixels in the original image between its center point and the center of a ground truth box is less than half of its diagonal or 100 pixels, whichever is larger.
In terms of the third dimension, the ground truth bounding box is assumed to span 25% of volume slices before and after the ground truth center slice and the predicted box center slice is required to be included in this range to be considered a true positive."
Thank you for your question.
The predicted center point is required to be within a *ground truth box* radius (or 100px) to be counted as a true positive.
"A predicted box is going to be counted as a true positive if the distance in pixels in the original image between its center point and the center of a ground truth box is less than half of its << of a ground truth box and not predicted >> diagonal or 100 pixels, whichever is larger."
We will clarify the definition of true positive predictions.
MateuszPosted by: mateuszbuda @ Dec. 30, 2020, 11:29 p.m.
Thanks for the clarification. Now that makes sense.
Another question. Is the objective of this challenge to detect the biopsy proven cancer lesions only, or it is to detect any of those lesions that underwent biopsy (doesn't matter whether it turns out benign or cancer)?
The goal for this challenge is to detect all biopsied lesions (both benign and cancers).Posted by: mateuszbuda @ Jan. 7, 2021, 8:01 p.m.
So do we even need to classify them into cancer and benign?Posted by: quanchen @ Jan. 7, 2021, 10:22 p.m.
You are correct; there is no need to try to classify your detections as benign or malignant. The score that's mentioned in the instructions is the confidence level (as estimated by your method) that a detection represents a biopsy-proven lesion.Posted by: kdrukker @ Jan. 8, 2021, 12:28 a.m.