Is 5-axis IBIS only a marketing ploy?

There are six degrees of freedom, as illustrated by the movements of a ship:

Three of these movements are rotations, which corresponds to the movements of a camera on a fixed tripod whose position is being adjusted by moving the tripod head: tilt, pan and rotation between landscape and portrait mode.

The other three of these movements are translations, which correspond to bodily moving the camera through space while it remains pointing in the same direction (as on a gimbal, for example) - fore/aft, left/right, and up/down.

Camera-based stabilization, be it in the lens, via a moving sensor, or by a gimbal, can only correct for rotations. It cannot correct for translations, which is why so many people end up frustrated with their attempts to shoot video while walking - although the camera is stabilized, the up/down bobbing that happens while you walk is often plainly visible. It's also why it's difficult for stabilization to handle very close subjects like macro photographs - at close distances even small translational movements of the camera appear as fairly large excursions of the image in the frame.

Full stabilization in translation basically requires a tripod or similar fixed mount. For video, you often want to stabilize two translation axes while moving in a third, which requires more specialized equipment such as sliders, dolly tracks, cranes, or a full steadicam rig.

I wonder what pixelshift has to say about it. Pulling the center shaft of the tripod down certainly looks very different with stabe on or off, so I wonder...

This is known as a "pedestal move" in the cinematography world.

If everything in the image frame is at the same distance and relatively far away then the change in point of view won't be very obvious.

But if there is a foreground vs. a background then when you drop the camera the foreground objects will appear to rise faster/higher than the background objects. Traditional camera- or gimbal-based stabilization can't prevent that. In fact nothing can prevent it if your aim is to move the camera down to a lower position. Even if you tilt the camera up as you drop it so that the foreground object appears stationary in the frame, the background will appear to drop.

Do you know of any measurement that shows jitter with IBIS compared to shooting on a tripod or with a fast enough shutter speed?

This is actually a theorem, requiring some efforts to prove!

Actually, those are combinations of rotations and translations. Also, this depends on where you put the origin of the coordinate system.

Pointing at the same direction is not sufficient to characterize translations. The camera might be rotating with respect to its axis.

Actually it can fully correct for only one of them.

Sure, lateral translations change the perspective.

Of course there can be compound movement. But for the purpose this discussion of in-camera stabilization (where the coordinate system being used for compensation must necessarily originate at the camera itself), there's a qualitative difference between translations and rotations in terms of the motions that can be compensated for.

If that were true then a camera that's simultaneously tilting and panning would end up with horizontal or vertical blur. Yet actual stabilization systems routinely compensate for this.

You could note that simultaneous tilt and pan can be redefined as a single rotation around a different axis, but IMHO that's adding unnecessary complexity to the discussion.

blog.kasson.com/?s=ibis

The second sentence of yours does not relate to the quote above. Anyway, you want the origin to be in what I believe is called a nodal point. Going back to what the camera can fully compensate for: only for rotations about the axis, nothing else.

Or with geometric distortions.

Speaking about IBIS - only to some extent.

Thanks. I do not see any measurements where IBIS reduces IQ at high shutter speeds.

Many ML bodies have 5 axis correction.

I find Nikon 5 axis IBIS combined with in lens VR (which requires a Z lens with in lens VR) can be more effective than in lens VR with an F mount with in lens VR unit.

How much of the improvement is due to greater in lens VR capability in recent lenses is something I cannot measure.

I just accept improved performance 😀

A Nikon specific detail - on a Nikon 5 axis body with pre Z lenses only 3 axis correction is available.

There are a few, but the delta is slight, and possibly within the margin of error. The Sony a7R and the Sony medium long zoom is an exception, because of shutter shock.

I find no difference, in the case of the 70-200/2.8 E and S.

Yeah, at least for strictly mechanical IS systems. Worth noting, though, that cameras like the GoPro which perform stabilization through software manipulation of the image can correct for geometric distortions because the software algorithm knows the lens profile and can correct those distortions based on the inputs from the camera's motion sensors.

I don't have a machine that can hold a camera with high precision and shake it and the lens the same way repeatedly with different bodies, and I don't know of anyone else doing it, to compare the negatives of stability. I have my field experience, and the testimony of many other people in Canon forums who felt that they lost consistence of stability when switching from a Canon DSLR to the Canon R5 with the same long lens with OIS, over a range of practical shutter speeds. The only plausible explanation is that the added IBIS is making errors, either sloppily or by correcting things that don't need to be corrected. The extra blur is very small, in the range of 0.5 to 1.0 pixels with the 45MP FF sensor, but it happens in the entire frame, so it is not coming from sensor rotation, but from sensor translation. The only reason an IBIS system should be translating the sensor with a non-hybrid OIS lens is to correct for detected camera/lens translation, so I suspect slop and/or overcorrection. I have considered that perhaps the correction is quantized, which could cause corrections to be larger than the motion, for very small body translations. Obviously, if one really needed large body translation correction, of many pixels in length, quantization errors would be lost in the large benefit. If the IBIS corrected a simple linear translation motion of 7.8 pixels with a correction of 8.2 pixels, it would obviously be a net benefit of tremendous value, with a net blur of 0.4 pixels instead of 7.8, but a motion of 0.2 pixels that is corrected with 0.6 pixels, however, would be a net loss of recorded stability, because it turned a 0.2 pixel motion into a 0.4 pixel blur. This seems to have improved with firmware updates, but it just goes to show how things can deteriorate in some cases while improving CIPA ratings. If we look at body/lens motions on a spectrum of blur sizes, then it is possible for large blurs to be very well corrected, while also adding shorter apparent motions.

One real-world example of correcting things that don't need to be corrected: I have two sets of Sony noise-cancelling headphones, one that cost $130 MSRP a few years ago, and another $349. The cancellation is much better in the $349 headphones, both by almost perfect cancellation of base frequencies, and by also cancelling at much higher frequencies than the cheaper set. However, the building I live in has a heating system where the pipes "hum" at a very low frequency, and I can't hear them with the cheap headphones, but with the $349 headphones, as soon as there is no audio to be heard, I hear the pipes humming. It appears that the headphones are correcting something that I can't hear in such a way that I can hear it. I've taken both on airplanes, and while the $349 headphones cancel engine noise much better, it reacts to rapid changes in air pressure in the plane when the pilot is making changes in airflow before takeoff that have an effect a lot like someone bludgeoning me with a rubber baseball bat. I can't hear the change in pressure at all with the cheaper headphones.

... which would imply that without a reported magnification from the lens for the focus distance, translation correction is never attempted?

and you decide to generalize that towards IBIS in general ?