i made new sampler just like libra sampling in rbbox_head with using rbbox_base_sampler.py and i got lower ap
seperatly i used orginal libra sampling in bbox_head but i got lower ap too
seperatly i used orginal BalancedL1Loss in head and got lower ap--------why SmoothL1Loss works better??
why libra sampling and its loss doesnt work better???
These methods may work well on some datasets, while for other datasets, adjusting highper-params are neccesary. I sugguest you to deeply analyse why these trick may help.
i made new sampler just like libra sampling in
rbbox_head
with usingrbbox_base_sampler.py
and i got lower ap seperatly i used orginal libra sampling inbbox_head
but i got lower ap tooseperatly i used orginal
BalancedL1Loss
in head and got lower ap--------why SmoothL1Loss works better??why libra sampling and its loss doesnt work better???
tnx for answering @csuhan