Sigma 18-35mm Autofocus: A Second Look

An analysis of focus issues on Pentax cameras

By PF Staff in Review Announcements on Feb 2, 2015

The Sigma 18-35mm F1.8 "Art" is an impressive lens.  Not only is it the world's first DSLR lens with a fixed aperture of F1.8, but it has also proved to be an optical masterpiece with image quality and bokeh comparable to that of high-end prime lenses.

Unfortunately, in our in-depth review of the 18-35mm, our staff reviewer Heie discovered that the lens suffered from serious autofocus issues while shooting through the viewfinder.  These initial findings were verified using five copies of the lens, at different times, across three different K-3 bodies among others.  We therefore suspected a problem with how the lens interacts with the camera during phase detect autofocus (PDAF), as no issues were observed in the live view contrast detect autofocus (CDAF).

However, while many user reports agreed with our findings, other users stated that they haven't experienced any autofocus issues whatsoever.  Thus, in an attempt to arrive at a more conclusive verdict regarding the 18-35mm's autofocus, our forum administrators decided to test two additional copies of the lens side-by-side.  Read on for their findings!

The Problem

While shooting through the viewfinder, the 18-35mm can often fail to focus accurately on subjects that should normally not pose an issue for a modern camera system.  The degree of inaccuracy can be so great that frequently leads to unusable photos in practice.

Our original findings suggested that the 18-35mm suffered from front- and back-focus so inconsistent that it could not be reliably corrected via the Sigma USB dock.  In practice, this meant that many photos shot in everyday scenarios ended up being blurry despite receiving AF confirmation from the camera.  Because the 18-35mm has a relatively low degree of magnification, AF inaccuracies can usually not be discerned through the viewfinder, and this can easily catch you off-guard.

Cameras & Settings

For the tests to follow, we used a Pentax K-3 with center-point AF.S PDAF in focus priority mode.  Initial tests were repeated with a second K-3 body as well as a K-50 to verify the results.  The aperture was set to F1.8.

Lens Firmware

We checked for firmware updates using the Sigma USB dock and were informed that firmware v1.00, the version installed on the 18-35mm from the factory, was the latest firmware.  We thus proceeded with testing.

Sigma USB Dock for Pentax

Baseline Test

To simulate a real-world shooting scenario, we started out with an informal test to see if either copy of the lens (henceforth referred to as copy A and copy B) could focus on a car parked some 20 meters from the lens.  At various focal lengths, both lenses failed to focus accurately even after multiple attempts.  We were able to get a few sharp images of closer subjects, albeit never consistently.

We then re-attempted the test in live view and got perfectly-sharp images every time.  This logic carries over for the remainder of the tests we performed: wherever viewfinder PDAF failed, live view CDAF was always able to deliver a sharp image.

Test Charts

Next, we proceeded to shoot an angled focus test chart at 18mm, 24mm, 28mm, and 35mm with both copy A and copy B.  The test chart was positioned less than a meter away from the camera, centered, and shot at an angle of 45 degrees.  20 photos were taken at each focal length for each lens: half from minimum focus, and half from infinity.  Thus, we ended up with 160 images.

The purpose of this test was to observe focusing errors rather than attempt to precisely quantify them, for which a LensAlign chart would be preferable. 

Table 1 shows the type of inaccuracy we observed at each focal length.

Table 1: type of AF inaccuracy

These categories correspond to the following marks on the test chart:

• Slight: focal point >15mm, <25mm
• Moderate: focal point >= 25mm, < 50mm
• Severe: focal point >= 50mm

Our test charts led to some interesting findings.  For any given focal length, each lens suffered from either front-focus or back-focus, but the type varied between copy A and copy B.  For example, at 18mm, copy A consistently exhibited front-focus, whereas copy B consistently exhibited back-focus. The degree of each inaccuracy was significant, but at any given focal length the type and severity was consistent for each lens, with the following exception.

More interestingly, the focusing error would intermittently end up being "off the charts" about 20% of the time during tests in which moderate or severe inaccuracies were otherwise observed. This resulted in a completely blurred front-focused or back-focused image as if the camera were focusing on something not visible in the frame.  Despite this, the camera would still report the image as in focus without making further attempts to adjust the focus.  This happened both when starting from infinity and minimum focus.  As you will see in the next section, it also ocurred in our test photos.

Overall, most of the test chart photos were slightly out of focus (i.e. within the limits of the test chart), none were perfectly in focus, and some were completely out of focus.

Test Photos

In order to test how these results translated into real-world performance, we kept the camera on its tripod and proceeded to photograph trees at close (2m), medium (10m), and long range (100m) with both lenses.

We captured 10 photos of each tree at the same four focal lengths as before and check to see which ones could pass as being in focus.  We only captured photos when the camera gave AF confirmation.   The results were less than thrilling; a 26% overall keeper rate when summing up the breakdown in Table 2:

Table 2: number of acceptably-sharp images

Figure 1: medium distance test scene

Below are two photos shot with copy A of the lens at 35mm and medium distance.  Image 1 is what we would consider to be acceptably sharp, while Image 2 is severely front-focused and thus unusable if one were trying to portray the cactus.  This is representative of other photos that ended up being "off the charts" in our prior test, and it was also a common ocurrence in this test as shown in Table 3. 

Table 3: number of "off the charts" images

Click on either thumbnail for the original image:

Image 1: sharp / slight FF

Image 2: blurry, "off the charts" FF

The fact that the lens regularly produced such blurry images while the camera reported AF lock is puzzling.  While Image 1 would appear perfectly sharp at a smaller aperture, there is little hope for recovering Image 2.

AF Corrections

We of course wanted to get the 18-35mm to start taking sharp photos, so the natural next step was for us to compensate for the front and back focus by applying AF adjustments via the Sigma USB dock. 

Let us start by taking a look at the Sigma Optimization Pro software interface:

Figure 2: Focus adjustments via USB dock

As shown in Figure 2, adjustments can be made at four focusing distances at four different focal lengths.  This yields a total of 16 possible inputs.  Each can be adjusted from -20 to +20 and defaults to 0.

Using the tree 100m away from us, we then proceeded to enter adjustments at infinity at 35mm for copy A of the lens.  After about half an hour of fine tuning and verification, including comparisons to CDAF results, we ended up with a +6 adjustment at infinity.  We re-shot 10 photos as in the previous test and to our surprise, nine out of ten ended up being sharp.  Great news!

While still zoomed in to 35mm, we then positioned objects at minimum focus, 35cm, and 50cm in order to fill in the remaining adjustment values.  None proved to call for adjustments greater than 10, and the adjustments seemed to sufficiently address the focusing issues at their respective shooting distances.  Again, great news.

The bad news, however, is that these efforts only yielded a lens that was capable of focusing reliably at the calibrated distances.  Objects farther away than half a meter (the third adjustment point) but not quite at infinity (i.e. the trees at 2m and 10m) continued to give the lens trouble and the adjustments had virtually no effect on the overall keeper rate following a second test.

Lastly, we redid the adjustment at infinity using the tree 10m away as our target and entered an adjustment of +9.  This in turn led to a 0% success rate at infinity but 100% success for the tree 10m away following a repeat of the earlier test.  In other words, it was possible to get one or the other consistently sharp, but never both.

Final Test

To wrap things up, we took the AF-adjusted copy A out for a quick photo shoot just to see the practical effect of our AF adjustments at 35mm.  Subjects between minimum focus and 20m away were photographed for a total of 19 images, but only 9 of those ended up being in focus.

Omissions

We did not attempt to calibrate copy B and we also did not perform AF adjustments at any focal length other than 35mm with copy A.  We also did not try using other AF settings such as AF.C, or global AF adjustments in conjunctions with the Sigma USB dock.  Investigating these additional paths could certainly prove interesting but this is beyond the scope of this article.

Conclusion

Given the aforementioned test results and the fact that we have now used a total of seven copies of this lens, we recommend that buyers exercise caution when considering the Pentax K-mount version of the Sigma 18-35mm. Having warranty coverage or the ability to return/exchange this particular lens is very important.

We believe that the lens suffers from autofocus issues caused by more than pure bad luck, as we were unable to get the lens to reliably focus in PDAF mode despite our efforts.  The camera would frequently report the image as in focus even when the degree of error was so significant that re-acquision should have been attempted.  However, we cannot identify the source of the Sigma 18-35mm's AF issues beyond speculation as we lack intimate knowledge of the technical implementation of SAFOX (the Pentax AF system) or Sigma's hardware.

One interesting thing we observed was that at times, the Sigma would overshoot the correct focal point and backtrack.  This behavior, with the exception of very fine adjustments, normally only happens in CDAF.  In the video below you will notice that it happened both while starting at minimum focus and at infinity in PDAF:

Video: Lens sometimes overshoots point of proper focus in PDAF

This overshooting often resulted in very blurry images when combined with the presence of FF/BF, but not always. Rather than trying to "hunt", the camera just gives up and reports proper focus. Conversely, the lack of an overshoot did not guarantee correctly-focused images. It could therefore be possible that the Sigma lens misinterprets the camera's commands or sends back improper signals with some degree of tolerance.

Another explanation could be the fact that a fast zoom lens simply does not play along well with SAFOX, but this is doubtful, as we experienced no issues with the Sigma 35mm F1.4 "Art".  Nevertheless, it cannot be ruled out since the Sigma 18-35mm lens is unique among zoom lenses with its F1.8 maximum aperture.

Since the Sigma 18-35mm exhibits no issues whatsoever while focusing with CDAF in live view, we can potentially rule out the HSM motor as the source of the PDAF issues.  The way in which the camera communicates with the lens (or vice-versa) is more likely to be the culprit, especially if the interface has been reverse-engineered by Sigma.

Our tests also offer a potential explanation for why some users haven't experienced any AF issues with this lens:

• due to potential sample variation, one could get lucky and obtain a copy without FF/BF at critical distances
• at narrow apertures, many subjects will appear acceptably sharp even with a moderate degree of FF/BF, i.e. Image 1 if it were re-taken at F2.8.

Finally, we have unmasked a potential shortcoming of the Sigma USB dock AF adjustment feature: it offers no adjustments between 0.5m and infinity for this particular lens, a range in which we would expect a significant number real-world photos.  It would be interesting to see if adjustments at a longer distance, such as 2m, would be possible and if they would in turn improve real world results.  To the best of our knowledge, every current Sigma lens can only be adjusted at four different shooting distances.  Those distances vary from lens to lens.

Many questions remain unanswered and the way this lens behaves during autofocus continues to be puzzling.  With the number of variables involved it is difficult to draw definitive conclusions through empirical testing. We therefore plan on bringing up this issue at CP+ 2015 in mid-February.

Last week, Sigma very efficiently addressed an SR-related problem with the Pentax version of its 17-70mm "Contemporary" lens by releasing a firmware update.  Thus, there is hope that we might see an update for the Sigma 18-35mm if its AF inconsistencies can be addressed via firmware.

For those still considering the Sigma 18-35mm, which is otherwise a very capable lens, don't miss our in-depth review.