Wheeler delayed choice experiment : why the mirror is not an interaction collapsing the wave?

I came across the Wheeler delayed choice as it is described in Wikipedia :

To summerize, the individual photon has many paths and some path leads to detectors that reveals the path taken and some path reunite the 2 paths and so we see the interference figure (I don’t get in the "delayed" part of the experiment because it is not the point of my question)

My question is that all paths involve mirrors (or "semi-mirrors") that change the direction of the photon. Why does this direction change is not an interaction between the photon and the mirror material that would result into collapsing the wave function ?

Is it because the photon is either fully deflected or fully pass through so there is no "exchange" with the mirror ? Meaning that the wave function can "bounce" on another material without having to "choose to be a particule" ?