IOU Validation metric ¶
By: mostafa.hisham1995 on Oct. 25, 2023, 1:52 p.m.
Hello all, I have just a simple issue\question. For semantic segmentation of the tool types, each tool is classified in a different frame\channel. As some data sets contatin only 2 or 3 types, some channels always have low value. If these channels are calculated by IOU it will have zero true positives which will lead to lower mean IOU. It is stated in the paper that frames not containing tools are ignore. but due to some inaccuracy in the training model, some False positives are resulted, but having no true positives and alot of labels the IOU will always be low, so can I ignore all channels that have zero positives in its ground truth or not. The purpose is to compare results with the paper