Open HKKTT opened 9 months ago
The AP_Single
reported in our paper is calculated by cocoAPI, not the print values following AP_Dual(Joint_AP).
You can refer the Table 4 of our paper in this link https://arxiv.org/pdf/2304.10765.pdf
But I looked at the source code of bodyhands, and it seems that this dual ap is not referring to joint ap; instead, it is more like single ap. I don't know if I misunderstood it.
The definition of AP_Dual
is a bit confusing. This indicator is newly defined by BodyHands. I just adopt its official code in my evaluation for fair comparing.
https://github.com/hnuzhy/BPJDet/blob/main/utils/eval/eval_bodyhands.py
You may consult the authors of BodyHands with your questions.
All right