Originally Posted by ericjut
...since wider lens require less stabilization.
Interesting question from a software standpoint: does it make a difference to the "shake canceling" algorithm whether a frame was shot in tele or wide? My guess is that your statement is operationally correct and it is what we all believe, pretty much. But in computing terms, if there are recognizable edges or other artifacts in the frames being stabilized, I suspect the algorithm is no more efficient on wide shots than tele shots. Data is data.
I think people tend to overcorrect for their movements when zoomed in because they can see they're off more easily - that is, the content of the frame changes more and it's easier to see that, so you do more sharp movements to correct this, and overall, you get lots more movement out of the human when zoomed in. I think that if a human had really tight movement control zoomed in, the algorithm would have no efficiency loss over the same human's filming at wide angle. Conversely, if someone makes big jerky movements filming wide angle, the algorithm will be no better at suppressing that than at suppressing big jerky movements zoomed in. If the algorithm can grab something in the content to use as a reference area for stabilization, it's good. People still have to avoid jerkiness regardless of their zoom level.
That said, the implication is that if you're filming something discrete that the algorithm can use to compare against all the time, it seems to be great at damping down your twisting motions. I suspect that if you film something so bland there's nothing discrete to grab onto, the algorithm gets defeated altogether. So that could be something to keep in mind when choosing your subject. If it has identifiable objects, a horizon line, or something similar, the EIS will be more effective.