I guess maybe CR allready does this, but just to be sure.
- From photo exif data, the focal distance can be read
-the focal distance from the camera in the reconstructed scene (depth area with sharpest image and most tie-points found) can be found - the ratio of the previous two variables could be used to scale the scene approximately to real-world sizes, without having to input measures manually?
Currently, my scenes seem to have rather random scale by default.