[Dipsoid]

Nor can I, Phase detect is done on a sensor that is completely separate from the imaging sensor. There is a possibility that IBIS can affect the precision of CDAF.

I could be wrong of coarse, but from what I've read, there might be some inaccuracy in the focus plane because the sensor might be shifted to a different coordinate than from what the phase-detect sensor "sees." To correct this, the phase-detect module would have to be stabilized as well, unless software processing calculates some sort of focus offset. Honestly from real-world use I highly doubt that this has much effect on picture-taking, if any at all.

[Digitalis]

what I've read, there might be some inaccuracy in the focus plane because the sensor might be shifted to a different coordinate than from what the phase-detect sensor "sees."

this probable would also affect optical image stabilization as the focal plane will also be shifted by optical means and will likewise be shifted during exposure - so neither approach is 100% sub-pixel precise.

[Fogel70]

I was watching Marc Levoy's lectures on Digital Photography, and in Lecture 10 he suggests that in-lens stabilisation is better for AF than sensor stabilisation in DSLRs - does anybody know anything about this, and if it is the case, could it be a factor in Pentax's lag in AF capability?

This is more of a theoretical problem than a real one. I never seen any scientific test done on this to prove that AF work better with optical stabilization.
For it to be a real problem you probably need to shake the camera more than any stabilization system can handle.

The only situation I would believe it can make a noticeable difference in normal use is if following a small moving subject using continuous AF using center AF point only.
But that can probably be called operator error more than anything else.

One thing I noticed though is that when the viewfinder is stabilized you cant't sense how much you shake the camera, so it basically impossible to focus on holding the camera as still as possible.
But with non-stabilized viewfinder it's easy to do that as you see the camera shake in the viewfinder.

[biz-engineer]

The AF integration time is even recorded in the EXIF data and ranges from 2 msec for bright high contrast subjects to 250 msec for dim, poor contrast subjects.

Interesting. Where do you find this data in the exif? Would you be able to post an example?

[nono]

this probable would also affect optical image stabilization as the focal plane will also be shifted by optical means and will likewise be shifted during exposure - so neither approach is 100% sub-pixel precise.

No, this would not as far as AF is involved. Basically, light flows in a single path through the lens to the mirror, where it splits in two paths, one to the main sensor, one to the AF sensor (actually, in a classic SLR, it doesn't as both sensors cannot be active at once because of the mirror, but the geometry of the body is equivalent; only the Sony SLTs are actually built this way AFAIK). In-body stabilization can make one of these path shorter or longer than the other as the main sensor moves, in-lens cannot.

[ffking]

Thanks for fascinating replies -I learn a lot from these forums.
As I understood it, in layman's terms - and I remain a total layman - the point made in the lecture (and Levoy seems largely ignorant of Pentax technology) is that with in-lens stabilisation the image is stabilised before it hits the AF sensor so the AF sensor has a more certain signal to work on and can therefore make its calculations quicker, thus quicker AF.
This is not a problem in mirrorless cameras, where IBIS is more common. Which did make me wonder if Pentax can ever catch up with Canikon in AF speed with its sensor based stabilisation, despite the all the other advantages the floating sensor enables.

[Digitalis]

In-body stabilization can make one of these path shorter or longer than the other as the main sensor moves, in-lens cannot.

Incorrect. The focal plane can shift with optical stabilization due to the IS elements trying to correct for the users movement. Can inadvertently cause tilt or shift at the focus plane in a similar way to how tilting or shifting the front standard of a large format camera. IBIS is already shifts the rear standard, the front standard remains static, optical stabilization essentially works the opposite way.

---

Last edited by Digitalis; 09-22-2016 at 03:57 PM.

[TER-OR]

There are advantages to both. If you like small lenses, IBIS has a lot of advantages. If you like vintage lenses, IBIS is the only option. For a large portion of your photography IBIS is an advantage and keeps lens prices lower. For long lenses, particularly for sports etc, ILIS holds an advantage. Note that the Sigma long zooms for Pentax are available with ILIS.

To have nice old lenses stabilized is really nice.

[photoptimist]

Interesting. Where do you find this data in the exif? Would you be able to post an example?

Here's some more technical links that mention this issue that I'm sure you will like:

LumoLabs

Pentax Tags

EXIF fields, Pentax-3 tag0x021f Makernotes Cameras by GVsoft

From the hints I've seen online, the camera's choice of "AF Integration Time" seems pretty complex because it not only has to deal with light level and contrast level variations but also avoid integration time values that would interact poorly with 50 Hz or 60Hz flicker in artificial light sources.

[stevebrot]

Interesting. Where do you find this data in the exif? Would you be able to post an example?

Pentax Tags

As with all makernotes tags, the names are the ExifTool author's best guesses as to purpose. They are "deemed accurate, but not reliable", to borrow the phrase used by realtors in the MLS book.


Steve

[biz-engineer]

Here's some more technical links that mention this issue that I'm sure you will like: LumoLabs Pentax Tags EXIF fields, Pentax-3 tag0x021f Makernotes Cameras by GVsoft From the hints I've seen online, the camera's choice of "AF Integration Time" seems pretty complex because it not only has to deal with light level and contrast level variations but also avoid integration time values that would interact poorly with 50 Hz or 60Hz flicker in artificial light sources.

Thanks for the links.

As with all makernotes tags, the names are the ExifTool author's best guesses as to purpose. They are "deemed accurate, but not reliable", to borrow the phrase used by realtors in the MLS book.

It is written: "values below 2ms are 0."

[Winder]

I was watching Marc Levoy's lectures on Digital Photography, and in Lecture 10 he suggests that in-lens stabilisation is better for AF than sensor stabilisation in DSLRs - does anybody know anything about this, and if it is the case, could it be a factor in Pentax's lag in AF capability?

With mirrorless cameras that might be true because the PDAF sensors are on the sensor, but with the DSLR, the PDAF is off sensor and not effected by the IBIS.

[RGlasel]

so the AF sensor has a more certain signal to work on and can therefore make its calculations quicker, thus quicker AF.

My understanding of AF sensors is that they only "see" a small strip of light and a basic histogram is compared to the other sensor strip at that AF point. Regardless, the AF sensors themselves don't do any calculations, the signals they produce are combined by a processor in the camera to estimate how long and in which direction the lens' AF motor should be operated. AF sensors produce continuously varying signals, which are read at points in time by the camera processors. There is no lag in the sensor, any delays in calculating instructions for the AF motor are produced by the camera's processors.

Phase detection AF in DSLRs doesn't involve the camera's image sensor, so moving the image sensor to stabilize the image (IBIS) has nothing whatsoever to do with AF. Moving the lens elements to stabilize the image (OIS) will affect the image detected by the AF sensor strips, but unless camera shake is affecting the lens motor instructions calculated by the camera, using optical stabilization doesn't aid or hinder AF either. Unless the AF system is precise enough to detect when the lens changes from being in focus to out of focus because of camera shake, OIS should have no impact on auto focus. The key in this is that image stabilization reduces changes to the image from camera shake while the image is being recorded, it doesn't bring the image into better focus. It is possible to take a blurry picture that is in perfect focus.

Contrast detection AF is a different beast and depending on how much of the sensor is compared each time, there could be situations where the camera gets confused by camera shake as to when the focus point is actually out of focus and when it is in a different position. In practice I doubt if this is an issue and even if it is relevant, it wouldn't matter if the image is stabilized in the lens or in the body, the degree of stabilization obtained would be the determining factor.

[Class A]

To be clear, IBIS does not affect PDAF.

The "problem" is, it does not aid PDAF.

As photoptimist correctly mentioned, PDAF sensors cannot "see" a scene instantly. They need some time to gather the signal (the darker it gets, the longer it takes). During that (so-called integration) time, the detail that the PDAF sensor sees gets smeared by motion blur, if there is camera shake. This blur makes it harder for the PDAF system to do the phase comparison required to calculate the amount of defocus.

An optical stabilisation approach avoids that motion-caused blurring.

BTW, there is a Pentax patent that suggests to shift the PDAF sensor based on data from the camera's accelerometers. The PDAF sensor would effectively move in tandem with the image sensor in order to counter-act the camera shake.

I don't know how much advantage an implementation of this patent would actually bring. I know for sure, though, that I wouldn't want optical image stabilisation for many, many reasons.

[photoptimist]

BTW, there is a Pentax patent that suggests to use the PDAF sensor based on data from the camera's accelerometers. The PDAF sensor would effectively move in tandem with the image sensor in order to counter-act the camera shake.

I don't know how much advantage an implementation of this patent would actually bring. I know for sure, though, that I wouldn't want optical image stabilisation for many, many reasons.

Interesting!

If they sample the PDAF sensor at a high rate (ensuring that each grab is relatively blur-free even if it's a bit dim) and then digitally average together enough frames to reduce the noise in the dim signal, they might be shifting the PDAF data digitally to compensate for camera motion.