Over the long weekend, I had the opportunity to run a series of tests with FoCal (Pro version). I'm running it on a 17" MacBook Pro (Core i5, 8 GB RAM) with the software installed on a Parallels vitrual machine running XP SP3, allocated one physical core (2 virtual cores) and 3 GB of RAM. I started with v1.0, and v1.1 was released over the weekend, so I updated mid-stream (no meaningful differences for me with the update).
My overall impression is very good, with a couple of caveats. The software is easy to use, although there were several random crashes (7 over a two day period). So far, I have tested 7 lenses on my 5DII (I may try a few on the 7D this evening). There are still a few bugs in the software, for example, it sometimes reports that the current adjustment is optimal, when looking at the detailed analysis data shows that a different one is best (supposed to be fixed in the next release). Also, while the testing was very straightforward with f/2.8 lenses and an f/4 lens, faster lenses gave it some trouble, sometimes resulting in a 'inconsistent results obtained so far' message (although allowing it to continue anyway often generated an excellent curve fit). It does have trouble estimating the distance (with the 5DII, at least), and I'm wondering if that's Canon's 'fault' or FoCal's? I know from the EXIF that Canon reports out two relevant values, 'focus distance upper' and 'focus distance lower' (not sure what these mean, if they define the DoF, or are confidence intervals, etc.). But for example, with the 135L it warned that the recommended distance is 6.8 m, and suggested that I was testing at ~8.1 m. Now, I'm pretty sure that 135mm x 50 = 6.75 m, and that converts to 266 inches, and I'm also pretty sure that I know how to use a tape measure - I doubt that I've managed to mis-measure by 4.5 feet! I'm not bothered by it, but it's worth noting. (If it is, in fact, Canon's 'fault' I wonder what that says about the distance information transmitted for E-TTL II flash metering…)
Setup was straightforward - the target search tool did a good job with most lenses/distances that I tested, although sometimes not, particularly with longer lenses at longer distances. The time taken for a test run was very short - even increasing the mirror lockup to 2 s (default is 1 s), test runs took less than 3 minutes. My setup consisted of the target taped to a wall, lit by three 150 W-equivalent gooseneck lamps pointed at the target from about 2' distance. The FoCal documentation recommends at least 10 EV, although it states that down to 5 EV will work; my lighting setup gave between 11 and 13 EV depending on lens and distance. Camera on tripod, with hotshoe bubble level to square it to the wall.
So, how has it worked out so far? Comparing the results to what I obtained manually using the LensAlign Pro for the same lens/focal length/distance, the results were very similar - spot on for fast primes, and within 1-2 units for slower zooms (where there are usually a couple of values that yield good results due to the deeper DoF). If I tested like I have with the LensAlign, the whole process would be much faster with FoCal. It takes me about 45 minutes to calibrate a lens manually (one distance/focal length, 30 minutes for setup, image capture, and tear down, and 15 minutes for comparisons); the same process with FoCal would probably take less than 10 minutes.
The advantage of FoCal is that it simplifies the procedure to the point where it's simple to test multiple distances and for a zoom lens, multiple focal lengths. Any particular adjustment you make is applicable for that focal length and that distance, and may not be optimal at other distances or zoom settings. With the exception of the 1D X, we're limited to applying a single adjustment value per lens - that means making the best compromise. The more data you have to help judge the best compromise, the better. Also, the 'best' adjustment may change over time, either due to a change in shooting habits, or a change in the camera itself (late last year, I mentioned that I dropped my 5DII a couple of times on an outing - I retested the AFMAs with the LensAlign after that, and found that for each of my lenses, the value that came out was systematically shifted 7-9 units negative from where it had been).
I tested at 50x the focal length (recommended by Canon and FoCal), 25x the focal length (recommended by LensAlign), and for the 100L Macro, I also tested near the MFD using the small focus target. Without moving furniture around, the longest line-of-sight inside my house is ~35 feet, so I didn't test anything longer than 200mm (for which I'd set up outside, but daytime temps over the weekend were down in the single digits!). For zoom lenses, I chose focal lengths loosely based on the zoom ratio, e.g. 3 points for a 2-3x zoom and 4 points for a 4x zoom.
Here are the tabulated results for the lenses I've tested so far, along with the AFMA value I selected for each lens.
Lens Focal length 50x AFMA 25x AFMA Selected AFMA EF 16-35mm f/2.8L II USM 16mm -5 -3 -3 24mm -1 -2 35mm -3 0 EF 24-105mm f/4L IS USM 24mm -3 0 -1 50mm -1 0 85mm -3 0 105mm -2 -1 EF 70-200mm f/2.8L IS II USM 70mm -7 -7 -6 135mm -8 -5 200mm -8 -3 EF 35mm f/1.4L USM 35mm -6 -11 -9 EF 85mm f/1.2L II USM 85mm -10 -9 -9 EF 100mm f/2.8L Macro IS USM 100mm -1 -2 -1 ~MFD: 2 or 0 or -1 EF 135mm f/2L USM 135mm -10 -12 -11
Several observations occurred to me when looking over the data. First, distance matters - in some cases, testing at 25x vs 50x makes a 5-unit difference in the optimal AFMA, which is more than half of the depth of focus for the lens at max aperture, i.e. will be noticeable in real-world shots. For zoom lenses, different focal lengths give different results, and the progression is not necessarily linear nor unidirectional for a given lens (so, I wonder how the 1D X will handle the two values for a zoom). I selecting an AFMA, I considered both the results at the different distances in terms of DoF for a given focal length, and also the distances which I commonly shoot for those lenses. For example, the value I chose for the 35L is biased toward the closer distance, since that's where DoF will be thinner; I picked the value for the 85L (even though there wasn't much difference) based on the fact that I usually shoot that lens at about 7-8 feet for portraits, which is ~25x the focal length.
Other notable findings not evident from the data were that the testing was quite consistent. Borrowing concepts for assay validation from work, I checked intra-run and inter-run variability, and both were quite low - repeating the same same lens/focal length/distance several times in succession yielded the same value (±1 for f/2.8-f/4 lenses), and re-testing after tear down and re-setup the next day yielded the same results. One exception was testing the 100L near the MFD, which gave more variable results; not a real problem since I almost always manually focus for macro anyway. One other interesting observation concerned the target itself - for the first lens I tested, I decided to repeat the LensAlign manual method on the spot for a direct comparison to FoCal. After the manual test, on a lark I connected the camera to my Mac and ran Focal with the camera pointed at the LensAlign. FoCal reported the same adjustment as it had with that lens using the printed FoCal target (which was the same as the manual value, too). Finally, the developer is quite responsive, as we've seen already from posts by Bob and others - good support is a critical part of new software!
I'll follow up once I have some data from the 7D and/or longer lenses.
Reply With Quote
