About using the target domain gt_boxes

[Zysty]

Hi ! Thanks for your share! In the filie HTCN/lib/model/faster_rcnn/faster_rcnn_HTCN.py, I found that you used the target gt_boxes in the RPN, resuluting calculating "d_ins" with target gt_boxes. While in the related code of "Strong-Weak Distribution Alignment for Adaptive Object Detection ", the related results are returned before RPN.
So I think this is not really UDA for detection... Is there a problem with this part？ Thanks for your reply!

[chaoqichen]

Hi @Zysty

Please refer to this issue for details. Note that we did not use the gt_boxes for training any classification or regression losses. Moreover, please download the latest codes.

Best, Chaoqi

[Zysty]

OK. Thanks for your quick response!