search

MengHao666 / Hand-BMC-pytorch

PyTorch reimplementation of hand-biomechanical-constraints (ECCV2020)
MIT License
69 stars 10 forks source link

Question about results of using BMC Loss in supervised 3d hand pose estimation #5

Open xxxshark opened 2 months ago

xxxshark commented 2 months ago

Hi, thanks for your work. But I have some problems about BMC Loss. I want to add BMC Loss in my experiment which estimates 3d hand pose from a single gray image. But, it's hard to converge. When I add BMC Loss, my original 3d keypoints loss will rise, so I have to decrease BMC's loss_weight util 1e-9. After training, I compare this experiment with BMC Loss and the baseline without BMC Loss. The baseline without BMC Loss performs better in MPJPE and it has lower BMC Loss in the same test dataset, which means BMC Loss has a negative effect in my experiment. Is it normal I have this conclusion? Or what maybe wrong in my experiment?

MengHao666 commented 2 months ago

I think you should first check if my joint order in computation is consisitent with yours. And the constraint range of mine is from several dataset. Also, the bmc loss is a weak-supervised loss and is proved to be effective in such weak-supervised setting.

xxxshark commented 2 months ago

Thanks for your reply. I check again and the joint order is same with yours. The constraint range of mine is calculated from my train dataset. The code in constraint range recalculation and loss calculation is same.