Sony In-Body-Image-Stabilization (IBIS) aka SteadyShot: Is There a Downside?
Related: digital sensor, Eastern Sierra, image stabilization, In-Body Image Stabilization, Matt G, Mt Dana area, peak bagging, Sony, Sony A7R, Sony A7R II, Sony A7R III, Sony mirrorless, Thorsten K, tripods and support, Zeiss, Zeiss Batis
Get Sony A7R II and Zeiss Batis B&H Photo.
I have this nagging feeling, based on field images, that IBIS (SteadyShot) might damage image sharpness, particular when the user is a bit wobbly in odd vectors (me, standing on a 30° slope with the wind blowing). I say this based on a few hundred images I took handheld (see Dana Glacier hike) which for some images have these weird blur characteristics that I am simply baffled by and unable to explain. And these blurs are not typical motion blur; they are varying asymmetric blurs and even look like rotational blur at times. Sometimes not there, sometimes subtle, but my eye has been trained by a few hundred K digital images over years. So I pay attention to pattern recognition by my noggin.
Then too is this idea: with IBIS = off, what guarantees plane parallelism of the sensor? Of course there has to be a lockdown mechanism, but just how accurate and precise is it? The “5 axis” IBIS feature moves the sensor freely, so how exactly is the sensor locked down to plane parallelism to within a micron or so? Perhaps it has more variance might be hoped? I ask this question in part because of the Sony 35mm f/1.4 Distagon skew.
Another question: if a lens is adapted (via lens adapter) to the A7R II (“dumb” adapter), what are the behavioral differences with that non-native lens versus a native (non-OSS) lens with IBIS, and a native lens with IBIS disabled?
I don’t have the answers for these questions, at least not yet. I’m hoping that Sony can fill me in what the behaviors ought to be.
The SteadyShot feature is described by Sony as follows:
The new flagship a7R II model is equipped with an innovative 5-axis image stabilization system that has been fine-tuned to support its high-resolution shooting capacity.
Similar to the system launched on the acclaimed a7 II model, this advanced form of image stabilization corrects camera shake along five axes during shooting, including angular shake (pitch and yaw) that tends to occur with a telephoto lens, shift shake (X and Y axes) which becomes noticeable as magnification increases, and rotational shake (roll) that often affects video recording. This camera shake compensation system is equivalent to shooting at a shutter speed approximately 4.5 steps faster.
Additionally, the 5-axis stabilization works cooperatively with Sony alpha lenses with optical SteadyShot™ (OSS) to provide maximum stabilization and clarity, while also performing admirably via a compatible mount adapter with Sony alpha A-mount lenses without on-board stabilization. Effects of the stabilization can be previewed via live-view on the LCD or OLED viewfinder of the camera.
Thorsten K writes:
You worry about sensor to mount parallelism with IBIS off.
No idea here, but a follow up question: I guess with IBIS being ON, the sensor would not be parallel by design, then? Since IBIS may tilt the sensor to counter-act the user moving the camera? Then for landscapes, IQ would be degraded by design when using IBIS (focal plane would be unpredictable). Or am I missing something?
DIGLLOYD: see notes from Matt G below, which are correct: the sensor is always plane-parallel to the lens flange. However, pitch and yaw are compensated for by moving the sensor in a way that is supposed to optimize the image, and this approximation can cause errors in some areas of the image while improving others. This I take as explaining the confusing sharpness results with the 25mm f/2 Distagon in the Dana Glacier examples. It also explains why my Zeiss Batis 25mm f/2 Distagon tripod-based shots with IBIS off look terrific—no funny business going on with the sensor planarity since no sensor movement.
As per my contact at Sony:
Moving the imager to create pitch/yaw compensation would take the sensor out of perpendicular alignment with the lens. Pitch and Yaw compensation are approximated by X/Y translation – I believe the same as other 5-axis sensor shift systems.
The only system of which I’m aware that can apply mechanical pitch and yaw compensation are those in which the lens and sensor move together as a locked system – Sony uses this in Balanced Optical Steadyshot, included in some fixed-lens camcorders.
Matt G writes:
You ask what guarantees plane parallelism of the sensor with IBIS off. The answer is that it is guaranteed mechanically - the sensor only ever moves in one plane, regardless of the motion being compensated for. This is a source of much confusion since camera marketing materials often suggest the opposite (the Olympus 5-axis system is exactly the same as the Sony one in this regard, despite Olympus publishing diagrams that appear to show the sensor tilting!)
The facts are that tilting the sensor would not successfully compensate for pitch or yaw, and would only serve to throw large parts of the image out of focus. You would have to tilt the lens and sensor together for this to work, but this would require lenses capable of tilting, and huge motors to make it work!
Mathematically speaking, pitch and yaw movements give rise to a projective transformation of the image (the sort of transform you get in Photoshop when using the "free transform" tool and moving the corners of the image). This transformation can be approximated by translation (moving the image up/down left/right), but the approximation is only valid for one location on the image plane, and the error increases with distance from this point.
You can demonstrate the problem easily using a tilt-shift lens. Set the camera up on a tripod with the lens unshifted, lining the centre focus point up with a strong image detail. Take a photo. Pan the camera on the tripod slightly, inducing a yaw movement. Next shift the lens so that the same image detail is again exactly under the central focus point. Take another photo. When you switch between the images the centre will look roughly the same, but the corners will appear to change shape.
This effect is what is causing the asymmetric blurs you are experiencing.
The problem is not with the Sony implementation, IBIS is not a substitute for preventing camera motion in the first place, and I would not use it for landscapes or still life images where the corners of the image will be in focus.
DIGLLOYD: well, I guess I don’t understand what “5 axis” means then. I can understand translation (left/right, up/down) and rotation, but that’s 2 directions plus angular for a total of three. But if pitch and yaw are compensated for by those movements by moving the sensor to some “intersection point”, then it seems that the cure is worse than the disease and I’d like a setting to disable pitch and yaw correction. I’d rather have traditional motion blur than some unpredictable asymmetric weirdness.
Matt G follows on to the above:
In response to your updated blog post, disabling pitch/yaw correction would be fairly pointless. Around 90% of camera shake (depending on focal length) is due to pitch and yaw. You might as well disable the entire system (which is something I would recommend).
Camera translation (up/down, left/right movement) has very little impact on the image (in general). If you put a telephoto lens on a tripod look at a distant subject and then drop the centre column by 1cm, the image in the viewfinder will hardly change at all. Pan the camera slightly and there will be a huge shift in the image. Translation only has a noticeable effect at high magnifications.
The second point in favour of pitch and yaw correction is that the longer the focal length, the closer the real projective transformation gets to a affine transformation that is readily approximated by translating and rotating the image. A simple way to visualize this is to again mount a telephoto lens and pan the camera left and right and the image seems to simply shift left and right. Now mount an ultrawide lens and do the same, the corners appear to stretch and warp in a complex way.
This is true of both optical and sensor-shift stabilization systems, and explains why image stabilization first appeared in telephoto lenses and is rare in wides.
IBIS ought to work fine with no artifacts for normal to long lenses (for small movements, otherwise you run out of sensor travel and image circle). But for something like the 25mm Batis I would only use it for shallow DOF shots where the corners are out of focus.
DIGLLOYD: I’m going to disable IBIS for wides in unstable situations or a lot of near/far detail, but use it for f/2 and f/2.8 for dusk shooting or similar. My tests show that at 1/20 sec, IBIS helps. I am dubious about 1/60 on up, where I think I can do better without it.
