In my view, the box_cls_map and centerness_map are the normalized score after sigmoid already, so
0<box_cls_map<1
0<centerness_map <1
, then,
0 < box_cls_map * centerness_map < 1,0 <= center_aware_weight * box_cls_map * centerness_map < center_aware_weight
, so,
atten_map >= 0.5 forever.
If we set center_awareweight to [0, 1], the
center_aware_weight * box_cls_map * centerness_map_ is also a normalized score,
then why does the atten_map need the sigmoid again?
Hi, your work is great.
I wonder why does the atten_map need another sigmoid at this line?
https://github.com/chengchunhsu/EveryPixelMatters/blob/fcb5703b347be7e44361efe4316e191076fd0d1f/fcos_core/modeling/discriminator/fcos_head_discriminator_CA.py#L70
In my view, the box_cls_map and centerness_map are the normalized score after sigmoid already, so 0<box_cls_map<1 0<centerness_map <1 , then, 0 < box_cls_map * centerness_map < 1, 0 <= center_aware_weight * box_cls_map * centerness_map < center_aware_weight , so, atten_map >= 0.5 forever.
If we set center_awareweight to [0, 1], the center_aware_weight * box_cls_map * centerness_map_ is also a normalized score,
then why does the atten_map need the sigmoid again?