in bifpn,why there are some connections with weigh,while some are not

[li-yun-peng]

in bifpn.
in add operation.
why there are some connections with weigh,while some are not,will that make a difference in training?
with
p6_up = self.conv6_up(self.swish(weight[0] * p6_in + weight[1] * self.p6_upsample(p7_in)))
without
p7_up = self.conv7_up(self.swish(p7_in + self.p7_upsample(p8_in)))