NVlabs / PoseCNN-PyTorch

PyTorch implementation of the PoseCNN framework

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Unexpected losses

hrishikeshkale4 opened this issue · comments

I am running the script ycb_video_train.sh and I am getting only the loss corresponding to the (these are fields printed on the terminal) 'label' and 'center' as non zero. But the losses corresponding to 'box','loc' and 'pose' are zero in the very first few iterations. Does that mean something is going wrong because losses shouldn't be zero in the first iteration right ?

Could you please help me. Any help in this regard is highly appreciated

Thanks
Hrishikesh

There is no loss for box and pose in the beginning because the network can not detect the objects yet.