how to use piou loss in Mask RCNN-based methods?

[ustczhouyu]

Hello, I want to replace Piou loss with Smoothl1 loss in the Mask rcnn-based instance segmentation method. I want to confirm whether I only need to replace Smoothl1 loss in RPN and Roi head with Piou loss? In addition, batch['cat_spec_mask'] in ctdet_angle.py is the segmentation label of the instance? Looking forward for your response, thank you.

[2018loveyyq]

Hello,and I want to know if you have solved this problem replaceing Piou loss loss with Smoothl1 loss in the Mask rcnn-based instance segmentation method??