[audiobomber]

I would love to see Pentax release a really nice K-mount mirrorless camera, which would do away with most of the PDAF pitfalls.

CDAF has its own issues, particularly with AF-C and tracking. The best Mirrorless focusses use PDAF for AF-C.

[Edgar_in_Indy]

CDAF has its own issues, particularly with AF-C and tracking. The best Mirrorless focusses use PDAF for AF-C.

I was referring to the shortcomings of the traditional PDAF systems as used in DSLRs, which have the PDAF sensors completely separate from the image sensor. But you're right, mirrorless cameras with the phase detection sensors on the image sensor kind of combine the best of both worlds... super reliable accuracy of CDAF with the blazing speed of PDAF. Recent mirrorless cameras with hybrid AF systems like the Samsung NX-1, for example, are actually terrific at tracking.

But anyway, we were talking about the 18-35mm 1.8, pretty much the antithesis of an action lens, so AF-C and tracking were not really what I had in mind.

[Stagnant]

In this thread I described my experience with the lens in question as an action lens. To sum it up : it can work as such.

[Digitalis]

mirrorless cameras with the phase detection sensors on the image sensor kind of combine the best of both worlds... super reliable accuracy of CDAF with the blazing speed of PDAF.

The only problem with on sensor PDAF sensors is that the homogeneity of the sensors surface architecture isn't even and will be visible under certain circumstances - the consequence of this is that you will get artifacts like this showing up in your images:



i'm quite content with dedicated PDAF sensors staying off the imaging sensor thank you very much.

---

Last edited by Digitalis; 12-04-2015 at 04:39 AM.

[Edgar_in_Indy]

The only problem with on sensor PDAF sensors is that the homogeneity of the sensors surface architecture isn't even and will be visible under certain circumstances, and you will get artifacts like this showing up in your images.

I've wondered before how that works, since it seems like having other sensors on the surface of the image sensor would either block or displace some image sensor sites. But I don't know enough about image sensors to know if that even makes sense.

Those two examples do not look good though. Do you happen to know how prevalent that issue is? Can you link to any full-size images where the effect can be viewed? I've read several detailed reviews of hybrid cameras in the past, and this is the first I've heard of this issue.

EDIT: I just went onto DPR and pixel-peeped several sample pictures from the Samsung NX1 and the Olympus EM1, and I couldn't find any artifacts like that. I couldn't find it in there A7RII photos either.

[Digitalis]

Do you happen to know how prevalent that issue is?

It is most clearly seen in scenes where subjects are lit with strong backlighting. It is also visible around strong light sources.

Can you link to any full-size images where the effect can be viewed?

I don't have any on hand myself. But it has sparked a heated debate in the Sony forum on the site that shall not be named.

I've read several detailed reviews of hybrid cameras in the past, and this is the first I've heard of this issue.

I was expecting something like this to show up eventually. Having read the white paper on Canons Dual pixel architecture, Canon evaded this issue because the homogenity of the sensors surface is maintained with the dual pixel architecture. The Sony approach does affect the sensor surface by the use of multiple phase detecting arrays in single pixel strips that run the length of the sensor.

[sholtzma]

I think there are two possible problems here at work:
[LIST=1][*]PDAF sensors cannot see how the focus actually is at f/1.8. Older cameras like K-5 have f/5.6 AF sensors, newer like K-5 II and K-3 have f/2.8 sensors (at least in the centre). If the lens in question exhibits focus shift (point of perfect focus shifts when stopping down) the camera has to guess where the focus will be at 1.8. The LensROM may(!) contain focus adjustment offsets that tell the camera in which direction and how far to compensate the focus to work around this problem based on the aperture used. I am not entirely sure, but I think if such a problem were present it would show itself in wrong (but consistent) focus.

Your idea intrigued me. Could our camera bodies be having trouble with f1.8 lenses, such as the Sigma? So, I was interested to run some tests (I've just begun) to see how well my copy of the 18-35mm performs as aperture changes. I chose a bright blue trashcan about 10-12 feet away, with large white writing on it, something easy to focus on. I set up my K-5iis on a tripod with 2 sec delay and took a series of shots using 18mm, then 24mm, and then 35mm. In most cases I refocused between shots; a few times I kept existing focus between shots. At each focal length, I took a shot at f5.6 and at f8; and for each of these two shots, I first focused using PDAF (2 shots), then CFAF using LiveView (2 shots), then manual focus (2 shots). Then, staying at 35mm (for easiest focusing), taking single shots, I opened up to f1.8, then to f2.5, then to f2.8, then to f4, then again to f2.8, f2.5, f2.2, f2, and f1.8. At f2 and f1.8 I took a shot using PDAF and another using CDAF. Then using PDAF I kept f1.8 and went from 35mm to 24mm to 18mm.

Results: in the series at f5.6 and f8, generally the camera focused well using any of the three focusing methods. There were one or two shots that were soft, either due to focusing inconsistency or perhaps my error. Interestingly, moving from f8 to f1.8 produced the worst focused shot of all. And focus stayed poor (though a bit better) from f1.8 through f2.8. But at f4 it was very sharp. Here I wondered if your idea was correct. But then I went from f4 down the list to f1.8, and shots stayed sharp down to f2. The PDAF shot at f1.8 was out of focus, but the CDAF shot at f1.8 was in focus. The last few f1.8 shots using different focal lengths were all in focus.

I'll try another set of shots later to exclude some more possibilities. For example, I'll refocus between each shot, I'll try only PDAF going up and down the apertures, etc.

[Nicolas06]

Your idea intrigued me. Could our camera bodies be having trouble with f1.8 lenses, such as the Sigma? So, I was interested to run some tests (I've just begun) to see how well my copy of the 18-35mm performs as aperture changes. I chose a bright blue trashcan about 10-12 feet away, with large white writing on it, something easy to focus on. I set up my K-5iis on a tripod with 2 sec delay and took a series of shots using 18mm, then 24mm, and then 35mm. In most cases I refocused between shots; a few times I kept existing focus between shots. At each focal length, I took a shot at f5.6 and at f8; and for each of these two shots, I first focused using PDAF (2 shots), then CFAF using LiveView (2 shots), then manual focus (2 shots). Then, staying at 35mm (for easiest focusing), taking single shots, I opened up to f1.8, then to f2.5, then to f2.8, then to f4, then again to f2.8, f2.5, f2.2, f2, and f1.8. At f2 and f1.8 I took a shot using PDAF and another using CDAF. Then using PDAF I kept f1.8 and went from 35mm to 24mm to 18mm.

Results: in the series at f5.6 and f8, generally the camera focused well using any of the three focusing methods. There were one or two shots that were soft, either due to focusing inconsistency or perhaps my error. Interestingly, moving from f8 to f1.8 produced the worst focused shot of all. And focus stayed poor (though a bit better) from f1.8 through f2.8. But at f4 it was very sharp. Here I wondered if your idea was correct. But then I went from f4 down the list to f1.8, and shots stayed sharp down to f2. The PDAF shot at f1.8 was out of focus, but the CDAF shot at f1.8 was in focus. The last few f1.8 shots using different focal lengths were all in focus.

I'll try another set of shots later to exclude some more possibilities. For example, I'll refocus between each shot, I'll try only PDAF going up and down the apertures, etc.

Focussing is always done wide open, it is only when the picture is taken that the lens is stopped down, so you would to experiment with lenses with different max appertures.

[Nicolas06]

Well, in theory the camera body is the one in charge of the AF. It tells the lens where to focus. The PDAF sensor in the camera body is supposed to report when focus is achieved. In the case of the 18-35mm, I'm not sure how the PDAF system could randomly be mis-reporting that the image is in focus, but apparently there is some kind of problem somewhere.

Typically for fast AF the camera compute where the focus should be put and give an order to the lens to move to the right position. If the computation is wrong or not precise, this can be the camera fault. This can be the lens fault if it lack contrast wide open and confuse the AF sensor. Then the camera give an order or activate the in body AF motor. If the order is wrong or the motor is moved too much or not enough, that's the camera fault. But if the in lens motor doesn't execute the order accordingly that would be the lens fault. If the mechanism inside the lens are not that precise and there a play or if the focus doesnt move as expected then this can be the lens fault.

CDAF solve the issue because it is an open feedback loop, focus is corrected until is it great, but that make it slow.

[sholtzma]

Focussing is always done wide open, it is only when the picture is taken that the lens is stopped down, so you would to experiment with lenses with different max appertures.

Got it. Thanks. So a potentially helpful comparison would be with the Pentax DA* 55mm at f1.8?

[Digitalis]

18mm, then 24mm, and then 35mm. In most cases I refocused between shots

This highlights one of the bizarre properties of this lens: when you zoom in the focus will shift, when you zoom out the focus will remain almost spot on, the focus shift when zooming out becomes undetectable at f/4.

[sholtzma]

This highlights one of the bizarre properties of this lens: when you zoom in the focus will shift, when you zoom out the focus will remain almost spot on, the focus shift when zooming out becomes undetectable at f/4.

Outwardly parfocal??

[Edgar_in_Indy]

CDAF solve the issue because it is an open feedback loop, focus is corrected until is it great, but that make it slow.

We should probably qualify the word "slow" here. CDAF may not be as quick as the fastest PDAF implementations, but it can still be pretty dang fast. In the latest mirrorless cameras, CDAF is extremely fast. I don't have the numbers handy, but it seems like when I've seen objective speed tests of various cameras, I've been underwhelmed by the speed differences between the best mirrorless cameras and PDAF.

And supposedly some recent mirrorless cameras can best DSLR's in low-light AF ability, focusing in very dim conditions where PDAF fails.

In real world terms, most mirrorless cameras are probably plenty fast enough for most people. My Panasonic GX7 is very fast, but even some older mirrorless cameras I've owned, such as my old Olympus PEN E-P1 were still pretty quick. Anyway, for the vast majority of my shooting I would happily take the accuracy advantage of mirrorless over the theoretical speed advantage of PDAF.

[Digitalis]

Outwardly parfocal??

At apertures of f/4 and smaller.

[Nicolas06]

Got it. Thanks. So a potentially helpful comparison would be with the Pentax DA* 55mm at f1.8?

DA*55 is f/1.4 wide open so that would be different but this would not be that fair because there much less dof even at same apperture at 55mm than 35mm. That the key point to me that got you the impression the focus was accurate once closed down: the dof is wider so the area that you selected is in focus even if it is not technically the part that is the center of the focus due to lens innacuracies.