Originally posted by raider
I just performed a focus chart test last night and would like to share my experiences.
I recently took some shots at a night concert using tamron 28-75 and fa50, and a lot of people commented that my shots are front focused. So I did a focus check using the famous D70 test chart.
I have Tamrom 28-75/f2.8, DA50-200 and FA50/f1.4. Since the tamron is my walkabout lens, I tend to focus on this lens first. Using the test chart and as per the author's instructions, I find that the lens is very much front focused indeed.
I then proceeded to set the "focus corr" value via the K100D debug service menu. After many trials and errors, I find that I need to set it to -220. A more detail test confirmed that this is incorrect. I need to set it to -350!! Now that is not a typo.
Then I tried the FA50/1.4, I need to set it to -80.
Then I tried the DA50-200. Due to the small aperture, the depth of field is much greater than the tamron and FA50, so the focus chart test is non-conclusive.
Now, K100D service menu can only have one focus corr value and out of desperation, I set it to -350 to suit my tamron. I thought this is all well but when I actually use the Tamron in real life shooting, the focus run back focus so much that it is not usable!
Put the k100D and tamron on the test chart again, the combo performed perfectly with -350 compensation. So I was freaked.
Then I decided to forego the test chart and adjust the focus corr myself. I set it to -100, -150, -200, etc and test and test. All no good. Finally, I set it to -50 and tested it!
Focus is spot on! With -50 focus corr, I also tested the FA50 and again spot on. This is the final value I settled with and I am not going to calibrate further.
My experience is that the test chart may not be that accurate afterall.
Of course, other people's experiences may differ.
IMHO, testing for focus accuracy is a difficult thing.
First of all, the AF sensor baseline corresponds to f5.6 so when someone is testing a f2.8 lens @f2.8, a slight front or back focus could happen (but it would be rather random) but the camera would still be within specs (let alone if you test a f1.4 lens!).
Then comes the question of light. I think it was RH (credit where credit is due) that outlined in one of his famous DPR post that AF precision might be altered by the type of light used. Now, in a perfect world, with a fully apochromatic light path, this could not happen since all wavelength should focus at the same plane. But in reality, different wavelengths will focus at slightly different planes so if you use a strongly monochromatic light (I'm thinking tungsten light here which is very cold, with almost no blue in the spectrum) and non APO lens, you could get some systematic shift in the focus plane.
There can also be methodology problems, do you focus on infinity first, then on your target or do multiple refocus on the target until you think you got it? Try it, you might be surprised...
Do you place your target @50x the focal length of the lens tested or at a fixed distance no matter what lens you're testing...?
All this to say that in the vast majority of cases, people are trying to judge accuracy of an AF with a 50mm f1.4 @1.4 under tungsten light and find strange results .... as they should!
Bottomline, no AF system is currently designed to give you perfect AF @f1.4 under any kind of light.
The best AF systems (ie Canon XXD/XD series, Nikon DX series and now Sony A700 it seems) are designed to give you "perfect" AF @f2.8, using the center sensor,and in white light but even them will eventually fail in artificial light or @f1.4.
In these case, there is no substitute for a good finder and manual focus.