Question about patch initialization

[nnop]

Hi, I have a question about the initialization step of PMVS.
In the following, I'll use I for image and f for feature.

The algorithm loop for all combinations of feature pairs fullfilling the epipolar constraint.
So if (I1, f1) and (I2, f2) created a patch when loop features in image I1, there would also be a feature created by (I2, f2) and (I1, f1) when loop features in image I2.
That way, there would be two times of duplicated patches being initialized.
Would that be problematic? Would the duplicated patches being filtered in the filtering step?