Importance of rpn_ bbox_inside_weights ?

[anand]

Hello Folks,

I was trying to work on object detection network and I found that for calculation of SmoothL1Loss (BBox loss), we need three input data: target, bbox_prediction,  bbox_inside_weights but I am not able to understand importance of  bbox_inside_weights parameter in SmoothL1Loss calculation

Can anyone help me to understand its role. It would be really appreciated.

Thank you.

Regards,

Anand