Here's the camera data taken from the LightStage knight's head data set (source http://gl.ict.usc.edu/Data/LightStage/)
#focal length
24745.1 24731.5
#principal point
1962.85 1167.7
#distortion coeffs
1.19791 2.1775 -0.0168257 -0.0151422
#camera tranformation matrix - Camera to World
0.99767 0.0598546 0.0327442 6.13211
-0.0584816 0.99743 -0.0413955 -13.0434
-0.0351378 0.0393841 0.998606 433.477
0 0 0 1
So the main issue is the 4X4 matrix, although the focal length seems odd and ideas on using the principal point and the distortion coefficients would be welcome too.
I tried plugging the matrix into decomposeMatrix.mel
http://www.sigillarium.com/blog/lang/en/186/
but didn't get a camera that seemed usable, quite possibly because I'm still in the wrong coordinate system.
Any ideas?
