It may sound a little bit specific, but…
It would be really nice if RC could handle planar mirrors in the reconstructed space. The way RC (and all photogrammetry software) handle the mirrors by reconstructing the space “on the other side” is quite fascinating but sometimes raises some issues (like big holes in the geometry because the space and the reflected space fight together).
The idea would be that the user put in space an approximate position and normal (as well as the mirror opening (its shape), maybe in the world space). RC would then, starting from this position and normal (plane of symetry), look for reflected tie points and use them for alignment, and possibly also for reconstruction and texturing.
I am often scanning artworks and sculptures that are in front of a mirror, and this could even help to reconstruct the back side of objects, that are well visible in the mirror, but not so in the space because of the angle.
Thanks !
Huh, interesting thought.
Might be a tad difficult to handle though.
What about covering the mirror for front shots and only use it for the difficult back side without anything outside the mirror?
Yes, very interesting, but :
-in the kind of Digital heritage scannings I am doing, covering the mirrors is not an option. We are not allowed to touch anything. 
-does RC handles mirror reflections ? if one of the cameras are taken from a mirror image, does it work ? is this handled by the lens distorsion coefficients ?
Oh, those hertitage guardians!
Same with flash photography - the effect is ridiculously small but still everybody gets up in arms…
Did you try masking the tricky parts?
Or you bring your own mirro - all new and straight and shiny.
Obviously on a hover platform…
Interesting question about the mirror distortion.
I guess it would only work if it was some kind of regular geometry.
But I might be wrong.
Are they really noticeably wonky, though?
Otherwise I wouldn’t worry too much - better some kind of back than none! 
Masking is a huge work (projects are typically 1000-3000 pictures and sometimes more).
Mirrors are quite good for the age they have, so they are optically very good.
But I am wondering if RC’s transformations could handle this.
I guess that it is easy to try with a few images normal anf a few left-right inverted. And see if every images are aligned together.
Yes, trying is always good! 
But even so, if the mirrors are good, what problem do you expect?
You would just need to mirror them back and the result should not be different from a normal image - especially in close range.
Well, it works !
Just adding the same pictures horizontally flipped did the trick and aligned with the others, allowing me to recover the backside of objects. Of course it comes with some issues, the main one being that the 2 reconstructed spaces overlap together if the mirror is “inside one of the spaces” (not strictly on the border).