SPIE-AAPM-NCI DAIR Digital Breast Tomosynthesis Lesion Detection Challenge (DBTex) - Phase 1 Forum

Go back to competition Back to thread list Post in this thread

> Number of bounding box examples vs training examples


boxes-train.csv contains bounding boxes for a lot fewer training examples than in file_paths_train.csv. Does this mean that the rest of the images do not have bounding boxes (no biopsy detected lesions)? It seems there are much less bounding boxes provided than the 700 scans for training.

Thanks you!

Posted by: marabian @ Dec. 27, 2020, 3:51 a.m.

yes, you are correct. Scans without bounding boxes do not contain biopsy-proven lesions. For any detection task, it is important to include 'normals' in the training/testing/evaluation. Hope this helps!

Posted by: kdrukker @ Dec. 28, 2020, 4:54 p.m.
Post in this thread