Originally posted by Eruditass This was a misprint from a K10D document, only(?) the K-7 has rotational compensation, the rest are translational.
Beware! The sensor can indeed translate (and rotate on the K7), but this does not mean it can compensate for translations of the camera!
In fact, only rotations of the camera can be somewhat compensated, translations cannot, as it depends on others parameters (unknown by the camera).
Thinking of it, rotations can only be partially compensated, as you then fall into the optical problem met in panoramic shoots where you have to rotate your whole system around the lens' nodal point, else you'll have parallax problems...
Thankfully, we're talking about really small angular displacements here, so they can be handled by a mere translation of the sensor, but this is an approximation.
In fact, the sensor-based SR implementation would have been better suited for compensating translations of the camera, but those don't have any real impact on the pic except for macro work (and you would have to somehow input the focus distance and magnification anyway!).
Now, regarding the possible firmware fix (one can always dream) : as they cannot possibly do a full study of all lens/grip/flash combinations out there, could it be possible to derive the needed SR countermove from the gyros readings?
Not for the actual shutter shake (as it seems impossible), but rather to have a pretty rough idea of how the camera is handled, and then adjust the shutter-shake pre-recorded removal pattern accordingly and send it to the SR module...
If the gyros are extra-still, then you can conclude we are on a super-strong tripod: Shutter-shake removal pattern for this shutter speed should be disabled.
If they are all over the place, then you have a light lens and/or a bad holding hand: Shutter-shake removal pattern for this shutter speed should be at full power...
If they are moderate, then you either have a bad tripod, or an heavy lens/flash/grip... The removal pattern would still be needed, but not as much...
With accurate mathematical modeling, I'm pretty sure they can pull this off... And the results would certainly be better than nothing.