Comments on MFA

[neuroanatomist]

It would seem logical to do it this way, in the field you could just set up on any point, do a focus and then a 10x check just to make sure you were hitting your target.

Upon further reflection during an otherwise relatively boring flight home to Boston after a day-trip to New Jersey (business - I wouldn't go there by choice...), I think I see a flaw in the logic.


I guess the limitation on the Live View method is the resolution of the LCD on the camera. With this method, you are in large part judging best focus by the sharpness of the 10x magnified image on the camera display. How many times does something look nice and sharp even when zoomed all the way in on the on-board LCD, but when you transfer the image to your computer you find that it's a little off. That's why the 'unofficial but Chuck Westfall-recommended so as official as unofficial gets' procedure for AFMA involves viewing images at 100% on your computer to judge sharpness.


I think that's an advantage to a tool like the LensAlign - you're setting AFMA based on a DoF scale, so that the plane of focus is centered on anangled rulerthat'sprecision-aligned to the focus target. I find that my quick reviews on the LCD (to make sure I've got the optimal adjustment bracketed within +10 to -10) usually give an estimate that's close to the final chosen setting (but not always exact, usuallywithin 1 or 2 units). The ruler means you're partly judging by sharpness, but really by the location of the region of sharpness as it moves along the ruler from setting to setting, as opposed to within-shot sharpness. Of course, after selecting and applying an AFMA to a lens, I always re-check the sharpness of AF shots in a 'real-world' setting (often my daughters eyelashes) - but again, that's checking at 100% on the computer.


--John

[Jon Ruyle]

I guess the limitation on the Live View method is the resolution of the LCD on the camera. With this method, you are in large part judging best focus by the sharpness of the 10x magnified image on the camera display. How many times does something look nice and sharp even when zoomed all the way in on the on-board LCD, but when you transfer the image to your computer you find that it's a little off.

Is 10x 1-1? If so, why is it less sharp than a computer screen? If not, I wonder why they don't just make it display 1-1?


Then again, isn't the 5DII screen supposed to have 900,000 dots or something? So at 10x, it should be able to display a 90 megapixel image at 1-1. (I know not the whole screen is used, but more than 25% of it, surely). Is 10x magnified beyond 1-1? It doesn't look like it to me.


What am I missing?


(Sorry for going a bit off topic here... I seem to have that habit)

[HDNitehawk]

Is 10x 1-1? If so, why is it less sharp than a computer screen? If not, I wonder why they don't just make it display 1-1?


Then again, isn't the 5DII screen supposed to have 900,000 dots or something? So at 10x, it should be able to display a 90 megapixel image at 1-1. (I know not the whole screen is used, but more than 25% of it, surely). Is 10x magnified beyond 1-1? It doesn't look like it to me.


What am I missing?


(Sorry for going a bit off topic here... I seem to have that habit)

To follow up on my reply to John earlier, and comment on this one. I do notice that if you are viewing a picture you have taken, sometimes it can look very sharp on 10x on the camera and still when it gets on the computer its just not as sharp.


But when focusing in 10x live view I have found it to be very accurate.

[Colin500]

But when focusing in 10x live view I have found it to be very accurate.

Same here, at least with the right target — when using the manual "point at something and use live-view at 10x" method to perform the AFMA, the target was an image containing very fine grids, which amplifies very small focus changes by the changes in the interference/moiree patterns in the live-view screen.


Ciao, Colin

[HDNitehawk]

[quote user="neuroanatomist"]I guess the limitation on the Live View method is the resolution of the LCD on the camera. With this method, you are in large part judging best focus by the sharpness of the 10x magnified image on the camera display. How many times does something look nice and sharp even when zoomed all the way in on the on-board LCD, but when you transfer the image to your computer you find that it

[neuroanatomist]

Some good food for thought here, gentlemen...I'll try to serve up some more. [8-|]
<div>

Then again, isn't the 5DII screen supposed to have 900,000 dots or something? Is 10x magnified beyond 1-1? It doesn't look like it to me

</div>


Yes, 10x is magnified beyond 1:1. The specs for the 5DII and 7D list LCDs with "Pixels:Approx. 920,000 dots (VGA)" although on their 5DII specs page, Canon leaves out the word 'dots' which makes it even more misleading than leaving it in. Nowhere do they tell you that dots &ne; pixels, although it's implied by 'VGA'. In fact, they count each red, blue, and green subpixel as a 'dot' so they are calculating display resolution as VGA x 3, i.e. 640 x 480 x 3 = 921,600 dots.


Bottom line is that the display is 640x480, and since the aspect ratio of the LCD on these cameras is not 3:2, the image fills the width but not the height. So with a 7D's 5184 pixel width viewed at 10x, those 518 pixels are interpolated up to 640 pixels, i.e. 1.23:1, and with the 5DII's 5616 pixel width, viewing at 10x means 1.14:1. I'm not sure what algorithms Canon uses for the interpolation, but regardless, upscaling is bad for sharpness.


I don't think that viewing on the computer is strictly necessarywhen doing the AFMA, since if you're applying stepwise adjustments, you're not looking for the bestabsolutesharpness (for which you'd need to use the computer, due to the upscaling mentioned above), but rather the bestrelativesharpness across the range of adjustments (and the relative best should not depend on the viewing platform, provided the display is sufficient to allow you to resolve the differences; but that might not be the case...).

when using the manual "point at something and use live-view at 10x" method to perform the AFMA, the target was an image containing very fine grids

I've read about using moire-inducing patterns displayed on a computer screen for this as well. I did try that this morning, and found that with 10x Live View of a moire-inducing pattern on my laptop, I could move the camera back and forth a substantial distance without much change to the moire patterns (or at least the changes were too subtle to discern on the camera's LCD).


The other issue with that method (and with focusing on any image as a target) is the alignment of the target with the camera's sensor. Ideally, you want the target to becompletely flat andperfectly parallel to the sensor. In his description of how he shoots the ISO 12233 crops, Bryan mentions that, "Thecamera/lens is multiple-laser-aligned to the target..." That's really the main function of the LensAlign tool (without lasers, of course) - the sighting gates that allow you to align the target to the camera. If not for that, I could just prop a ruler against a wall and have saved a few bucks.

I ended up at +12, I found I could set the mfa doing it the live view way but it was just as tedious because you could almost see an acceptable range over 10 settings.

Hmmmm...I've never seen a range that large. There are several possible reasons for that. It could be the difference in method (Live View + AF vs. just AF and shoot), or to the different resolution and smaller size of the camera LCD relative to a computer display, or it could just be that your 24-70mm is really bad. [:P]


Actually, I think it's the difference in method - not just the Live View issue but the fact mentioned above - I'm using an alignment tool with a DoF scale so I'm not relying primarily on overall sharpness to judge the range.

Actually when doing Macro work I find the 10x method to be the most accurate method of getting focus and it can be very accurate

I agree, it's the most accurate. But consistent with your response to Jon, it's important to note thatviewing an image at 10x on the LCD to judge sharpness is not the same thing as focusing using 10x Live View, at least not for me. When I focus using Live View, it's an active, iterative process with visual feedback (just like the camera's autofocus, only I'm a lot slower) - I move the ring back and forth a few times to make sure I'm centered on the region where I want critical focus. If you AF on a point then switch to 10x Live View and see if it's sharp, that's static viewing; if you then rack the focus to check if the AF hit the spot, how do you know you're ending up at the same point? I think the only way to know for sure is to actually take the shot with AF then take the shot with MF 10x Live View, and view them on a larger display. Not just one shot, either. AF accuracy should be normally distributed(in the statistical sense) around the point of best focus - but a normal distribution doesn't mean spot on that best point every time. So you still need to take multiple shots.


I think it would be a bad idea to take the shortcut - i.e., out in the field, do an AF, switch to10x Live View and look, see if it's ok, if not then apply some adjustment and do it again until it looks right. If you're going to rely on AF, you want it as accurate as possible and setting it based on n=1 and/or shooting some random feature in the field is risky. I think that's why Canon recommends doing it only if necessary, and warns that it may prevent correct focusing from being achieved (presumably a result of doing the adjustment incorrectly, although the instruction manual doesn't actually say how to select the correct adjustment).


Ultimately I think the best way to judge is not to compareAF vs. 10x Live View MFon a single shot (or without a shot at all), but rather to apply stepwise adjustments and compare the results over several shots. Once you factor in the need to do that multiple times at each adjustment setting, I'm not sure that using AF and 10x Live View MF offers any time savings for AFMA, compared to just shooting the target with AF at a range of adjustment settings. That's especially true if you want to view the images on the computer.

[HDNitehawk]

or it could just be that your 24-70mm is really bad

John


I am of the opinion nowthat may be the case.


When I was testing the lens with Live View last night I would set it up and do a series of 10 at each spot, working my way up 1 each time. It was much quicker to just take a series of ten pictures and move to the next setting. A big disadvantage I see with a live view method is that you loose the ability to review the DOF accuratly. You can only really tell if the focus was dead on. Still, if you were in a situation that you had to do it this way, I think it would work to get by. The teadious method you described earlier I believe would deffinitly be the most accurate.


So far my set up has only been a little more advanced than Bryan's Ice Cream Box he used to calibrate lenses in his review. He recomended the "Datacolor SpyderLensCal Focus Calibration Tool". Do you think there is any advantage in getting a diffrent product. In the video the Datacolor Tool looked very small.


Rick

[neuroanatomist]

So far my set up has only been a little more advanced than Bryan's Ice Cream Box he used to calibrate lenses in his review. He recomended the "Datacolor SpyderLensCal Focus Calibration Tool". Do you think there is any advantage in getting a diffrent product. In the video the Datacolor Tool looked very small.

I started with a simpler setup, too, but a little more than an ice cream box. [] I used a printed test chart where you cut out and folded a target so you could focus on something flat at a 45&deg; angle to the printed ruler. It was a little kludgy, but it did convince me that I had some AF adjustment issues on my T1i, and that was one factor in my decision to upgrade to the 7D.


Datacolor has some nice stuff - I use and like their SpyderCube. TheSpyderLensCal looks convenient in that it folds flat, which the LensAlign does not. The flat target/angled ruler concept is the same, but I think the LensAlign offers two advantages.


The first is alignment - theDatacolor product has a level, and you can level your camera (hotshoe bubble level or built-in electronic on the 7D), but then you also would want to translationally align them. The LensAlign has a second piece parallel to and behind the focus target, and that second piece has bullseye dots that you align through holes in the focus target, meaning you can align the camera to the target at any angle rather than only level and orthogonal. Note that here I'm talking about the LensAlign Pro, not the basic version which lacks the rear plate and uses a mirror instead. Actually, I went to check that and noticed they have a new Mark II version (of the older basic one) which has the rear plate and disassembles to store flat. It lacks the Enumerator, though, which I find useful (that's a set of sliders that let you record the adjustment you applied right in the shot, although you could use a Post-It note for that, or just look at the EXIF in DPP).


The second advantage is the ruler itself, or rather the availability of a longer one. The standard ruler that comes with the LensAlign appears to be about the same length as the one with theSpyderLensCal, and that's fine for most lenses. But in some cases it's not long enough (i.e. the DoF of the focal length/distance/aperture being tested is nearly as long or longer than the Z-dimension of the ruler, i.e. the base of the right triangle where the ruler is the hypotenuse). Looking at your list of lenses, you'd be fine with either product when testing at 25 x focal length (which is what LensAlign recommends). But there are situations where you'd want a longer ruler - for example, wide angle but slow lenses on FF, a 400mm f/5.6 lens on FF or if you wanted to test your supertelephoto primes at a distance grater than 25 x focal length, such as the 50 x focal length that Chuck Westfall recommends). For those cases, LensAlign has a Long Ruler Kit that magnetically attaches to the standard ruler (and includes a larger focus target that also magnetically attaches).

[Richard Lane]

...TheSpyderLensCal looks convenient in that it folds flat, which the LensAlign does not. The flat target/angled ruler concept is the same, but I think the LensAlign offers two advantages.
<div style="CLEAR: both"]</div>

A lot of good info here. Thanks Guys!


Later this month, LensAlign will be releasing a less expensive, MKII model ($80) thatfolds flat. I take it that they needed to compete with the less expensive SpyderLensCal.


http://www.lensalign.com/


http://www.rawworkflow.com/


Rich

[Jon Ruyle]

Bottom line is that the display is 640x480, and since the aspect ratio of the LCD on these cameras is not 3:2, the image fills the width but not the height. So with a 7D's 5184 pixel width viewed at 10x, those 518 pixels are interpolated up to 640 pixels, i.e. 1.23:1, and with the 5DII's 5616 pixel width, viewing at 10x means 1.14:1. I'm not sure what algorithms Canon uses for the interpolation, but regardless, upscaling is bad for sharpness.

I don't mean to complain, and I'm sure the Canon guys have their reasons... but I wonder why the heck they don't just make it view 1-1? I mean, the whole point of 10x is to judge sharpness. Why not design it so it does that as well as possible?