NMS ??

[HwaiHo-0552]

As shown in Evaluator.py, the row 114 to 120, TP is the highest class socre bbox, and other bbox are set as FP.
I think it may makes error in recall and precision.
i.e. there is pred bbox named A=['dog', 'class score=90%', 'x', 'y', 'w', 'h', IOU=95%],
another pred bbox named B_1=['dog', 'class score=96%', 'x', 'y', 'w', 'h', IOU=89%]
A and B_1 are all belongs to possitive sample.
if we set B_1 as TP, meanwhile set A as FP, this will lead error to recall and precision.
SO, should set NMS to save B_1 as one pred bbox, then compute TP FP TN FN ??

[rafaelpadilla]

Hi @digitalized-hfut

In this example, as you realized, if both detections (A and B_1) are provided, A will be FP and B_1 is TP.

That's why applying NMS or any other approach to eliminate such undesired detection is a task of the object detection model. It is not a task of the evaluator.

Hope that answers your doubt.

Regards,
Rafael

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[HwaiHo-0552]

Hi @rafaelpadilla

Got it. Thank you for your code.

best regards.