That's about half a mil (I still think in mils) which is pretty close to the limit of resolution of ordinary machine tools and mechanical measurements. At that resolution surface finish becomes a real concern (apparently what you are seeing) and you need to be careful to wipe away the bigger dust particles.
I think we should be astonished at what can be achieved with modern manufacturing, even in consumer products, instead of being, well, spoiled children.
theSuede wrote:
That discussion has been up several times, and I've also spoken to quite prominent people at both Canon and Nikon about it. The counter-argument (to "making an AF-adjust and calibration cycle" built into camera fw) is always that the few customers that f*ck things up will make so much noise that the function will run a risk of actually losing them money.
This is why lenses and bodies should be profiled on inspection and values written to the FW, the user should not have to deal with this in the first place.
I think alpa now offer a service for MF digital lenses where they will test and select good copies and re-sell them at a hiher price.
And to get some perspective on that shim thickness; in a 200mm lens with 2:1 telephoto power (fairly normal?)
+/- those measly 0.01mm (1/250") gives a focus shift of 10cm at 10m distance.
+/- 0.05mm gives +/-25cm at 10m distance...
Getting results better than this has nothing to do with mechanical quality, at least not from a production cost perspective. Cost increases extremely quick if you want to decrease tolerances.
Doing it electronically in production on the other hand means a full additional cycle of product handling - even if it could be automated. That's also an additional cost in both time (the most important parameter for all lenses not VERY expensive) and in equipment, resource and facility costs.
You could do it better, you could get lower tolerances from manufacturing. But are you (as being "customers") willing to pay for it, that's the question...
knower wrote:
I agree in part with you theSuede, but I don't see how a "set the current value as zero" can screw things up more or less than a AF fine tune.
This is such a stupid and simple thing to put in cameras.
G.
Sorry, I do not understand. Zero of what?
Focusing is about recognizing how distant is your subject. At what distance would you put your test chart (or whatever), and how could you be sure that the distance is exact with millimetric accuracy?
I see plenty of ways for screwing things up by telling your camera that the chart is 3m away, while it's actually 3.01m.
Not to mention that you should do that for several distances...
Hopefully contrast detect AF systems or on-chip phase detect (like fuji) will do away with this issue, this is probably the only time it is going to get resolved.
Albi86 you are not understanding because you didn't read my previous post, I assume.
What we usually do is acquiring critical focus through the LV.
Then we tune the AF fine tuning to get as close to that chart as possible, in terms of sharpness.
This is accomplished by shooting pictures of the charts and comparing them to the LV one.
A simple function in camera saying use the "current detected AF value" as zero reference point would basically solve the issue completely and give us perfect focus in within 2 minutes.
You acquire focus with your LV. The camera measures the focus through the AF sensors and applies the difference automatically.
Very easy to do, very useful.
Of course the target needs to be put at the proper distance, but this is true even for a simple AF tuning.
knower wrote:
Albi86 you are not understanding because you didn't read my previous post, I assume.
What we usually do is acquiring critical focus through the LV.
Then we tune the AF fine tuning to get as close to that chart as possible, in terms of sharpness.
This is accomplished by shooting pictures of the charts and comparing them to the LV one.
A simple function in camera saying use the "current detected AF value" as zero reference point would basically solve the issue completely and give us perfect focus in within 2 minutes.
You acquire focus with your LV. The camera measures the focus through the AF sensors and applies the difference automatically.
Very easy to do, very useful.
Of course the target needs to be put at the proper distance, but this is true even for a simple AF tuning.
Sorry, but I don't see how the camera could recognize if it has to move focus back or forth with just one shot. Not to mention things like fringing, loss of contrast, field curvature, vignetting, etc. that can easily affect the perception of sharpness and fool such a simple algorithm.
It just has to measure the difference between the detected AF value and the LV one that YOU have put in perfect focus and apply the offset accordingly as the default starting point.
The camera doesn't have to compare anything but measurements, one is yours, the other one is the one of the camera AF.
If the LV has a value of focus that is 5 units off compared the the standard AF detect, then it has to shift the focus point by 5 units and set this as 0 (which is what it does with the AF fine tuning).
There is nothing complicated about that and all the things you are listing don't have anything to do with it, sine is just a measurement.
knower wrote:
It just has to measure the difference between the detected AF value and the LV one that YOU have put in perfect focus and apply the offset accordingly as the default starting point.
The camera doesn't have to compare anything but measurements, one is yours, the other one is the one of the camera AF.
If the LV has a value of focus that is 5 units off compared the the standard AF detect, then it has to shift the focus point by 5 units and set this as 0 (which is what it does with the AF fine tuning).
There is nothing complicated about that and all the things you are listing don't have anything to do with it, sine is just a measurement.
I don't want to go on and on forever. However, I still think that your method is flawed because it implies the use of AF to solve a problem with AF. You'll never know what your camera will focus on in order to compare the setting with the one you set in LV. As Einstein used to say, you can't solve a problem in the same way that caused it.
AFMA imho absolutely requires comparison and interpretation of the results. It's a back&forth, double-check process. What manufacturers could do, imho, is to provide their own FoCal-like software for free when you buy a high-end camera or lens.
You need at least two measurements at different distances. One at infinity, and it must absolutely not be against a blue background (sky). Already here it starts to get complicated. Best color is mid-green, so a forest scenery at 1 mile distance would suffice.
Then you need one more reference point, closer to MFD, but not at MFD. This needs to be at the same color-contrast as the infinity reference.
PDAF needs the zero extension (infinity distance focus) to work properly. And it needs the closer distance to set focus throw ratio.
The amount of circumstances that could throw a DIY adjuster off target is not small, it is significant. Bear in mind that FM forums probably contains a lot of people from the top percentile of tech savy users, and I wouldn't put it past many here to get it wrong.
Albi if you setup your targets properly the results are consistent enough.
You are not using the AF to setup anything else than just checking how it is OFFSET compared to your own focusing.
The lens is calibrated at a safe distance with the targets as per any other calibration system. This ensures that the error is minimal and gives as accurate measurement as possible with this kind of setup.
The delta of the measurements can be reduced by doing multiple measurements.
You focus the target with the live view manually, this will give you a focusing distance.
You let the camera focus, this will give you another distance.
The camera compensates through the AF fine tuning to overcome the distance as much as it can by bringing the focusing distance closer through the adjustments.
The accuracy is basically the same as using the DOT tune calibration method which has been proven to be quite good.
This would at least give us a quick way of fine-tuning a lens, even if as fixed focal length, in a quicker way.
theSuede who are you answering to? It is not clear.
If it was for me, well, I respect all the people here and I am happy to be proven wrong if that's how it is.
I am not scared at all to put out my point of view or what I think because other people could prove me wrong, actually this would teach me something instead.
Knower, that was a general response, not directed at a single target.
I will not go through the entire AF chain here, as that would be quite OTT for this thread. But the general outlines are official, and if you want more - track down a few service manuals for your brand of preference - you need them for both a recent body, and a recent zoom.
But there are a few main parameters that PDAF needs to be able to predict corrective movements in the way it does. The two most important ones are infinity position and throw speed. And you can't get both from one measurement, since there's no way for the camera to know if the error stems from the camera body's phase calibration or the lens's focus position indicator. The throw speed per focus position is a curve that's hardware locked in the lens, but it still needs to be positioned correctly. And so does the focus position offset.
If the body is off then a calibration test-setup with just one distance, one reference, will throw the accuracy off for all OTHER distances. That's just the way the system works.
theSuede I know what you mean, but I was assuming that we were in the simpler case, when what is off is jut an offset between the sensor and the lens, in fact I was referring to the AF fine-tuning, not the body's phase calibration.
That's a whole different thing, obviously much more complicated.
But the AF-fineTuning could be setup as I wrote in much less time and would be much easier to have something like that.
The fine tuning of one distance and the absolute position are intertwined. You can't get one without the other, and expect it to work at other distances than the one you tested at.
That's kind of the main thing about predictive open-loop PDAF, and the only reason for it being faster than contrast detection AF...
theSuede wrote:
The fine tuning of one distance and the absolute position are intertwined. You can't get one without the other, and expect it to work at other distances than the one you tested at.
Absolutely true, but I think what he's getting at is we are already doing this imprecise, distance-specific, AF fine tuning any time someone does the dot-tune method, or goes out and shoots test charts and tweaks the AF adjustment value.
In many cases, we're just making the best of the situation, getting it better than the factory calibration, even if it's not perfect and not applicable to multiple focus distances. The feature proposed - where you manually focus something and set it as the "zero" value for the PDAF system - is not going to substitute for some kind of complex factory calibration process but it would simplify the rudimentary AF fine tuning that we're doing now.
If the camera is capable of performing PDAF, then it should be capable of allowing the user to set focus, then measuring the difference between its PDAF "in focus" value and the current value, either by an actual measurement or by statistical means - for instance, having the user manually focus 5 times, initiate PDAF each time, and averaging each focus throw together.
This is, again, a rather blunt tool to a complex problem but it's less blunt than the trial-and-error process we're currently stuck with. It would also allow you to fairly rapidly perform adjustments in the field.
