hereszsz/thundernet-tensorflow2.0

F rpn F sam

henbucuoshanghai opened this issue · 1 comments

henbucuoshanghai commented

The DNN get the F rpn,
But the rois in the F sam and using roi align ,to get the bbox
but the F rpn is not used ???????

Horikita123 commented

It's used by the following way

thundernet-tensorflow2.0/layers.py

Line 361 in cf399bc

sam_result = sam([rpn_result, re], training=False) 

cem = CEM()
sam = SAM()
rpn = RPN()
re = cem([C4, C5, x], training=False)

rpn_result, rpn_cls_score, rpn_cls_pred = rpn(re, training =False)

sam_result = sam([rpn_result, re], training=False)