I recently did a survey using a precalibrated camera. When processing in Photoscan, however, when I fix the calibration under Tools->Camera calibration the results are way worse than when I do not fix the calibration. The difference in error is about one order of magnitude.
I suspect this is because if you fix the calibration Photoscan has less degrees of freedom moving things around to obtain a proper fit. Thus, is it better to precalibrate and get large errors, or do not precalibrate and basically do not know what changed?
I just processed the last of the test flights and the result is again interesting. Not fixing calibration gives 1.9m X error, 2.3m Y error and 3.4m Z-error, but when I fix the calibration the errors become 4.6m, 4.5m and 56m. This decreases the flight altitude from 113m to 61m as well, and the fly path was set to be 120m, thus the first is much more likely, considering the terrain is about 7m at its highest elevation.