Auto-focus sensors: high precision vs. regular testing

[hulkster]

One thing mentioned earlier that might be interesting is to shoot the 70-200/2.8 with and without the 1.4TC, using the same center focus point. I.e. one you add the TC, you no longer can use the high-precision sensors.

If I understand your conclusions correctly, we should actually see more reliable autofocus (under low light) with the TC attached.

I don't think so. Adding a tele-converter adds new challenges to the auto-focus system. It's not going to magically focus better just because the high precision point is disabled.

I'm confused ... isn't your conclusion that in low light, not using the high precision points yields better AF?

I.e. if there was some way to manually disable the HP points (not changing anything else), that would be the definitive test. Since presumably only Canon can do that in their labs, I was tossing out the idea of using the TC as a way of effectively disabling the HP points, but keeping everything else consistent, including the AF point.

I'm well aware the type of testing you are doing takes a ton of work, so just tossing out an idea. Again, great job and appreciate your posts.

 

[mmullen]

Your target mainly has contrast in horizontal and vertical directions in the center. This is an ideal target for a vertical or horizontal sensor, but not for the diagonal orientation of the high-precision sensor. That the different sensors may perform differently for different targets is another factor you failed to consider in your conclusion.

Actually, I can appreciate that. That's why I ran the test on this target as well:





I wanted to insure that the results would be the same on a target without strong directional texture.

But keep in mind the AF system can focus on very fine detail. I doubt the AF system was focussing on the colored patterns on the beach towel. This is what the AF sensor would have seen (or a portion of this):





The AF sensors do not see in color. In any case, that's why I used multiple targets to insure the results would be consistent.

--
Mike Mullen

 

[mmullen]

I.e. if there was some way to manually disable the HP points (not changing anything else), that would be the definitive test. Since presumably only Canon can do that in their labs, I was tossing out the idea of using the TC as a way of effectively disabling the HP points, but keeping everything else consistent, including the AF point.

The problem here is that adding a teleconverter would change far more than simply which AF sensor was used.

The focal length would be different, the light would be more dim, the AF algorithms would be different (it would be necessary to use a reporting TC) and the light from the lens/teleconverter combination would be converging on the AF sensors from different angles compared to a bare lens.

That would be introducing so many unknowns it would make a terrible test.

--
Mike Mullen

 

[mmullen]

And wouldn't a better test subject be an extremely low contrast subject with plenty of light? Like say a textured all white wall? I don't think you need HP sensors to accurately focus on that towel.

It looks like you do not understand the concept of AF precision. Every system has it's limits of accuracy. No matter how easy the AF target is to read, the regular precision AF points will not return as high of precision as the high precision sensors.

In most cases the little bit of extra precision is not even necessary but it can be nice to have if you are shooting a wide aperture lens wide open. A good F4 lens will do fine with the precision of the regular AF sensors.

--
Mike Mullen

 

[hulkster]

I.e. if there was some way to manually disable the HP points (not changing anything else), that would be the definitive test. Since presumably only Canon can do that in their labs, I was tossing out the idea of using the TC as a way of effectively disabling the HP points, but keeping everything else consistent, including the AF point.

The problem here is that adding a teleconverter would change far more than simply which AF sensor was used.

The focal length would be different, the light would be more dim, the AF algorithms would be different (it would be necessary to use a reporting TC) and the light from the lens/teleconverter combination would be converging on the AF sensors from different angles compared to a bare lens.

That would be introducing so many unknowns it would make a terrible test.

--
Mike Mullen

OK - was just trying to think of some way to "fake" out the system for a test with/without HP sensors.

I do think your testing the center point and non-center point sensors is darn good ... and along those lines, I wonder if using spot focus might (effectively) disable the F/2.8 HP capability - from Canon's writeup, it says you still have cross-type sensors when spot focus is selected, but does not mention if the diagonal HP is still used.
http://www.usa.canon.com/dlc/controller?act=GetArticleAct&articleID=3049

 

[C S A]

Another method would be to set the aperture, pull the camera battery so the lens stays there, then tape over the aperture control contacts. I believe you can do that without affecting the AF ones.

A bit gimmicky, but it would fulfill the "same lens" requirement.

 

[mladavac]

Just out of curiosity, how did you reduce the illumination level? By dimming or shuttering the bulb?

Your target mainly has contrast in horizontal and vertical directions in the center. This is an ideal target for a vertical or horizontal sensor, but not for the diagonal orientation of the high-precision sensor. That the different sensors may perform differently for different targets is another factor you failed to consider in your conclusion.

Actually, I can appreciate that. That's why I ran the test on this target as well:





I wanted to insure that the results would be the same on a target without strong directional texture.

But keep in mind the AF system can focus on very fine detail. I doubt the AF system was focussing on the colored patterns on the beach towel. This is what the AF sensor would have seen (or a portion of this):





The AF sensors do not see in color. In any case, that's why I used multiple targets to insure the results would be consistent.

--
Mike Mullen

 

[JohnMaij]

Why not test both at F4 thereby eliminating errors from the shallow Dof. The 85mm will still start off using high precision and only fall-back to normal precision if AF failed.

I wanted to insure that the results would be the same on a target without strong directional texture.

--
Mike Mullen

--

 

[Erik Magnuson]

Actually, I can appreciate that. That's why I ran the test on this target as well:

An ISO 12233 chart would be ideal - it has a circular pattern in the center that would obviate any directional sensitivity.

But keep in mind the AF system can focus on very fine detail.

The pixel pitch of the different sensors would be an item to test and not assume.

I doubt the AF system was focussing on the colored patterns on the beach

There are always going to be several potential phase matches. I don't know Canon's algorithm but the AF systems typically choose the area with the best contrast.

The AF sensors do not see in color. In any case, that's why I used multiple targets to insure the results would be consistent.

Actually, that's one of the new features on the 7D. It has a dual layer system to sense red/green vs blue/green in an attempt to not be fooled by CA.

--
Erik

 

[mmullen]

An ISO 12233 chart would be ideal - it has a circular pattern in the center that would obviate any directional sensitivity.

I like to avoid testing on targets that are too far removed from items in the real world that you might actually focus on.

The pixel pitch of the different sensors would be an item to test and not assume.

Why do you think I'm assuming the AF can lock onto a fine pattern? I have tested it. I have also seen in Canon technical literature the published spec for the AF pixel pitch. It is not quite as fine as the 7D sensor but it is close.

There are always going to be several potential phase matches. I don't know Canon's algorithm but the AF systems typically choose the area with the best contrast.

It would make sense that it chooses the phase match with the best match, not necessarily the one with the most contrast but that is just a guess of mine. But the technical details of exactly how it works might be useful for designing tests to probe how it works but I am more interested in how to get the most out of the tool. One thing I've learned is that a good f4 lens can AF in low light just as adeptly as the latest and sharpest f2.8 lens.

It has a dual layer system to sense red/green vs blue/green in an attempt to not be fooled by CA.

You are right, I had forgotten about that. It must then apply a correction factor to compensate for the behavior of different colors. It's amazing this stuff works at all, let alone so accurately and quickly on so many different types of targets and under such a wide range of illumination.

--
Mike Mullen

 

[mmullen]

Why not test both at F4 thereby eliminating errors from the shallow Dof. The 85mm will still start off using high precision and only fall-back to normal precision if AF failed.

The first test was with both lenses at f4. Of course the AF happens when each lens is wide open but I wanted the final images to be taken at the same aperture so differences in the depth of field did not corrupt the results.

The second battery of tests only used one lens (the newest 70-200 f2.8 IS II) and I used it wide open to manimize the depth of field and maximize the difference caused by small differences in focus.

My understanding is that a lens of f2.8 or faster always starts with a regular precision AF sensor and then when focus is almost achieved the high precision AF sensor is given a chance to take a reading and fine-tune the focus position.

--
Mike Mullen

 

[tko]

All the other replies are good, lots of interesting thoughts. Has anyone considered focus shift with aperture?

http://toothwalker.org/optics/spherical.html

Makes some sense that a lens focused at F4.0 and shoot at F4.0 would be better than a lens focused at F2.8 and shoot at F4.0. The 50L F1.4 is a prime example of this.

 

[copejorg1]

My understanding is that a lens of f2.8 or faster always starts with a regular precision AF sensor and then when focus is almost achieved the high precision AF sensor is given a chance to take a reading and fine-tune the focus position.

--
Mike Mullen

Mike, do you remember where you might have come across that information? I've looked quite a bit on the internet for some sort of "official" technical explanation for that very thing, and come up dry.

But the results of your own tests seem to contradict that notion. In other words, if the center AF point really did work that way (focus first using f/5.6 sensor, then pass off to f/2.8 sensor for fine-tuning), the result would never be worse than using just an f/5.6 sensor alone -- possibly better, but never worse. But your test results indicate that the presence of the f/2.8 AF sensor is making things worse in low light. Indicating that if the lens is capable of activating the f/2.8 sensor, the camera might be using only that sensor, and never even attempting to use the f/5.6 sensor within the same AF point.

However, someone else brought up the notion of using the "spot" AF option to see if that changes anything, and I think that's an interesting question. Does that option use a reduced-size version of the center AF point's "X'-shaped f/2.8 sensor in that case, when a fast lens is mounted? Or does that option simply force the center AF point to revert to the same reduced-size f/5.6 cross sensor that would be used at any other AF point, or with a slower lens at the center point, even with a fast lens mounted?

Excellent thread, BTW -- well done.
--
Greg

 

[mmullen]

Makes some sense that a lens focused at F4.0 and shoot at F4.0 would be better than a lens focused at F2.8 and shoot at F4.0. The 50L F1.4 is a prime example of this.

It's not focus shift as the lens stopped down. My tests done at f2.8 (wide open) showed the effect even more strongly.

--
Mike Mullen

 

[mmullen]

My understanding is that a lens of f2.8 or faster always starts with a regular precision AF sensor and then when focus is almost achieved the high precision AF sensor is given a chance to take a reading and fine-tune the focus position.

Mike, do you remember where you might have come across that information? I've looked quite a bit on the internet for some sort of "official" technical explanation for that very thing, and come up dry.

I'm sorry, I don't. I do recall it was a statement directly from a Canon rep.

But the results of your own tests seem to contradict that notion. In other words, if the center AF point really did work that way (focus first using f/5.6 sensor, then pass off to f/2.8 sensor for fine-tuning), the result would never be worse than using just an f/5.6 sensor alone -- possibly better, but never worse.[ quote]

No. If the f2.8 sensor didn't have enough light to work reliably, it could cause the focus to be "touched up" in the wrong direction.

Aother possible explaination is that the center AF point (being the only AF point with two AF points in one location) is compromised somewhat by this unique situation. In otherwords, maybe the center (regular precision) AF sensor is not as adept as the stand-alone AF sensors.

However, someone else brought up the notion of using the "spot" AF option to see if that changes anything, and I think that's an interesting question. Does that option use a reduced-size version of the center AF point's "X'-shaped f/2.8 sensor in that case, when a fast lens is mounted? Or does that option simply force the center AF point to revert to the same reduced-size f/5.6 cross sensor that would be used at any other AF point, or with a slower lens at the center point, even with a fast lens mounted?

That would be worth exploring. However these types of tests take a lot of time because they involve analyzing enough exposures to extract statistically significant results. I imagine the spot AF is accomplished simply by ignoring half of the AF sensor pixels. If so, it's likely that Spot AF is achieved on the high precision sensors in the same manner as it is on the regular sensors.

--
Mike Mullen

 

[mbloof]

Interesting thread. I've often wondered why fast glass can at times be hit/miss on focus while slower glass (and seemingly the 5.6 sensors) nails the focus more often.

I can't see how the differences can be tested if all the sensors on a camera are high precision. A camera with a low precision sensor(s) that is not shadowed by a high precision one but also has at least 1 high precision one might be easier to test.

Manually selecting a LP focus point and then a HP one might work or possibly setting up the focus 'target' so that there is only contrast under the sensor under test (with a 2.8 or faster lens attached will it use the HP sensor only and give up if the HP one does not work or will it revert to a LP sensor for a try?)

--
(insert brag sheet here)
http://flickr.com/photos/mbloof
Technologist @ Large

• Mark0

 

[tko]

As lenses perform their AF wide open, you can't change that.

The photo is actually taken at the stopped down aperture. Lens doesn't stop down till the photo is taken.

So, focus is at F2.8, shot is taken at F4.0. If focus shifts, you will have a problem. Problem is still real, just trying to understand.

 

[Dale Buhanan]

As lenses perform their AF wide open, you can't change that.

The photo is actually taken at the stopped down aperture. Lens doesn't stop down till the photo is taken.

So, focus is at F2.8, shot is taken at F4.0. If focus shifts, you will have a problem. Problem is still real, just trying to understand.

The only lenses I know that exibit noticable focus shift are the Canon 50/1.2, and I heard that the Sigma 50/1.4 does a bit also. This is supposed to be due to their aspheric elements and wide aperatures. As far as I know, the lenses in question here do not exibit focus shift with aperature. And if they did, it would be a very second order effect compared to the data for missed focus and of no consequence. Do you have data to the contrary?

--
kind regards
Dale

 

[TilmanW]

the 85-graph might give an explanation as to why I have had a terrible time focussing with that lens with outer autofocus points in the past.

Thanks!

 

[tko]

I have no evidence, only one of many theories, but perhaps people are looking at these things more closely, and things that haven't been noticed, or were imperceptible w/lower resolutions, are now being discovered.

Certainly the 50L has a lot of focus shift. So do other fast lenses, like the Noctilux which is known for its quite severe focus shift. Some enlarging lenses ( EL Nikkor 50mm f2.8, for one)do have a focus shift when changing aperture.

It wouldn't be a shock if slower lens have some slight focus shift. We are talking about a very sensitive test here.