Which is GT among the annotation-1, annotation-2, annotation-3?

Hello,
I am new for KiTS datasets and trying to figure it out.
But I found in case0~299, that the each instance (ex. ‘kidney_instance-1_annotation’ ) in instances folder has 3 annotation nii.gz files
(ex. ‘kidney_instance-1_annotation-1’, ‘kidney_instance-1_annotation-2’, ‘kidney_instance-1_annotation-3’)
The 3 annotation files might be came from the KiTS21 dataset which labeled with the reviewing step, but I can’t know which is final GT for segmentation. Which is final GT for training?

The 3 annotations are done by three individuals, each with the same intent, so the expectation is that the best annotation would be some aggregation of the three. Majority voting is one approach that might be used for aggregating these labels, and that is what’s done to achieve the “aggregated_MAJ_seg.nii.gz” file for each case.

1 Like

Ah, I see, thank you!
So where can i find aggregated annotation files like aggregated_MAJ_seg.nii.gz in KiTs23 datasets?
I could do the merging myself, but I think it would be more accurate if there was already merged data in the dataset.

1 Like

Those are just one folder up from the instance labels. For example, kits21/data/case_00000/aggregated_MAJ_seg.nii.gz

1 Like

The aggregated_MAJ_seg.nii.gz files are not in KiT23, but in KiT21! Thank you.

1 Like