Overall image quality
The 18-35mm delivers such extraordinary results in studio testing that it barely seems believable. However in real-world shooting it is genuinely capable of delivering wonderfully detailed images when shot wide open, at least when it's properly focused. This is shown in the examples below, which were shot at each end of the zoom, then converted from RAW with sharpening turned off in ACR and a relatively modest small-radius unsharp applied in Photoshop to bring out fine detail (amount = 150, radius = 0.3).
Canon EOS 100D, RAW + ACR
|18mm F1.8||35mm F1.8|
|100% crop, lower centre||100% crop, centre|
|100% crop, lower right||100% crop, bottom right corner|
At 18mm sharpness remains impressively high out to the edge of the frame, although green/magenta chromatic aberration becomes quite visible. Sharpness visibly drops right in the extreme corners, but most of the time you'll lose much more sharpness to the limited depth of field at F1.8 than you will to lens aberrations.
At 35mm F1.8 things are slightly different- if you shoot a flat scene straight-on you'll see a gradual drop-off in sharpness right across the frame. But this mainly due to curvature of field, so if you shoot something three-dimensional then objects in the extreme corners will look sharp if they happen to be in focus. The lens shows practically no chromatic aberration at 35mm, either, which means that the images look very clean indeed. Overall, this kind of image quality is remarkable coming from an F1.8 zoom.
Autofocus accuracy and consistency
One real problem we had in real-world shooting with the 18-35mm was with autofocus. Depth of field is limited at F1.8 - especially at 35mm - which means that even slight misfocusing can become very visible if you look closely at your images. We shot with a range of Canon SLRs, from the entry-level EOS 100D to the top-of-the-line EOS 7D, and all had problems focusing absolutely correctly all the time. This was usually only obvious when shooting at apertures larger than F2.8. In general, we got slightly better results by using the central AF point (with recomposition) compared to using off-centre AF points, but this didn't eliminate focus errors entirely.
The example below is typical of the kind of focus problem we saw. These two images were shot just a second apart, but while the top one is in more-or-less perfect focus, the other is slightly off - the camera has actually focused somewhat in front of the statue. This type of misfocusing seems most prevalent with relatively low-contrast subjects, and is often enough to take the edge off the sharpness. This means it can be difficult to make the most of the lens's excellent optics.
Canon EOS 700D, 35mm F2, 1/4000 sec ISO 200
When pinning down what's going on here, we have to draw a distinction between focus accuracy and consistency. Point an SLR at the same subject and autofocus 10 times while looking through the optical viewfinder, and it'll probably give 10 slightly different results. How close it comes on average to focusing correctly defines the accuracy, whereas the spread of focus distances describes the consistency.
We're not going to delve in detail into why this happens, however it boils down to the fact that SLRs don't use the main image sensor itself for focusing (except when in live view), but instead a different sensor with its own light path. This arrangement can give rise to both random misfocusing, and more systematic focus errors. The latter can to some extent be fixed using the 'Autofocus Microadjustment' settings in many modern SLRs, and offered in a more advanced form by Sigma with its USB Dock and Optimization Pro software.
A consequence of all of this is that in principle you'll get the very best results at large apertures by focusing and shooting in live view. This has its own pitfalls though - for example when focusing manually with Nikon's DX SLRs you have to remember to set the aperture to F1.8 before entering live view, as they're incapable of adjusting the lens's diaphragm during viewing. If you initiate live view at a smaller aperture, then you risk getting misfocused images. Meanwhile Canon users may have to set the aperture to F1.8 and press the depth of field preview button to force the lens wide open for accurate manual focusing, when shooting in bright light.
We looked at whether the focus problems we saw from the 18-35mm could be mitigated by using autofocus microadjustments. With a Sigma USB Dock to hand, we set about determining and programming in a full set of autofocus microadjust parameters for all of Sigma's specified focal lengths and focus distances (18, 24, 28 and 35mm; infinity, 0.5m, 0.35m and 0.28m). This took several hours to set up, even with specialised focusing targets to hand. The screenshot from Sigma Optimization Pro below shows the adjustments we ended up with (note that every copy of the lens is expected to require different numbers).
This procedure certainly improved overall focus accuracy when shooting at the distances used for microadjustment. However these are fixed by the software, and there's no option to specifically correct any distance between infinity and 0.5m. Unfortunately though, the vast majority of subjects end up somewhere in between, and we found that the lens still had some problems with focus accuracy even when fully programmed as above. Knowing this, we'd probably configure the 'Infinity' position with a bias towards getting the most accurate focus at closer subject distances.
The other take-home message here is that, while our copy of the lens showed a tendency towards focusing in front of subject (and therefore required positive microadjust values), the correction needed at each setting was different. So using a global in-camera microadjust value would necessarily be a compromise - setting a value to correct any given focal length and subject distance could throw off the focus at other settings. We observed this in practice using the Canon EOS 7D's microadjust feature - it could only be set to give optimal focusing at one given focal length and subject distance.
Are we looking too closely?
One point implicit in all of these discussions is that we're examining images for pixel-level sharpness, which is a hugely demanding criterion. Images don't necessarily have to be this perfectly focused to be entirely acceptable when viewed on-screen or made into in anything other than very large prints. So while accurate autofocusing is highly desirable, slight misfocus doesn't necessarily spoil a photograph completely. This is where each individual photographer has to make their own decision about what they're prepared to accept, based on personal experience and preference.
However, we believe most potential buyers for this lens will be demanding users - either serious enthusiasts or professional photographers. And while we think that they'll be delighted with the image quality the lens delivers when properly focused, they won't generally find this kind of hit-and-miss focusing at F1.8 very acceptable; the lens just doesn't reliably deliver usable images every time. This doesn't mean that there's no point in buying the lens at all - its outright image quality at normal working apertures is reason enough to get one - but it does mean that we'd recommend testing before you buy, if at all possible. Of course we've only used the Canon mount version of the lens - it's possible other versions may work better.
Specific image quality issues
As always, our studio tests are backed up by taking hundreds of photographs with the lens across a range of subjects, and examining them in detail. This allows us to confirm our studio observations, and identify any other issues which don't show up in the tests. In real-world shooting the Sigma 18-35mm produces consistently excellent results - just as long as the camera can focus it accurately enough.
The 18-35mm doesn't show too much in the way of chromatic aberration, but it's visible if you go looking for it. Under different conditions it can show the two main types of CA, lateral and longitudinal.
Lateral chromatic aberration is visible as colour fringing towards the edges and corners of the frame, and as usual is strongest at wideangle. It tends to be most obvious when the lens is stopped down a bit, which brings the fringing into sharpest focus. It's relatively easy to correct in post-processing, but will be visible in most cameras' JPEG output (although many Nikon SLRs will automatically remove it).
The Great Court of the British Museum is something of a torture test for this type of CA, and in the uncorrected example below you can clearly see green/magenta fringing across much of the roof structure when viewing the image up-close. This is simple enough to correct in most RAW converters - here we're using Adobe Camera RAW's 'Remove Chromatic Aberration' control.
RAW + ACR, no correction
RAW + ACR, CA corrected
Canon EOS 100D, 18mm F5.6, 1/100sec ISO 100
100% crops, upper left
Longitudinal chromatic aberration can also be visible occasionally, as colour fringing around out-of-focus areas of the frame when shooting at large apertures. This tends to be strongest at the long end of the zoom, but is never very objectionable. The example below illustrates just about the worst case scenario, with red fringing in front of the plane of focus, and green fringing behind, accentuated by the grass being over-exposed. This type of CA is generally less straightforward to remove in post-processing, as it shows up differently in every image.
|100% crop in front of focus plane|
|Canon EOS 100D, 35mm F1.8||100% crop behind focus plane|
Overall though, this level of chromatic aberration is impressively low, especially considering that we're looking at an ultra-fast zoom. It's no worse than we'd expect to see from fast primes, either.
The 18-35mm does reasonably well with respect to flare, given its fast aperture and the sheer amount of glass used in its construction. In real-world use shooting in bright mid-summer sun we've not seen any obvious problems, but the usual caveats apply - it's essential to keep the front element clean, and always best to shoot with the hood in place too. The examples below give an idea of how the lens deals with the sun in or around the frame.
In the first set we're shooting at wideangle with the sun in the corner of the frame, and comparing what we see with lens shot wide open and then stopped right down to make flare patterns most defined. Here the results depend upon the aperture; image contrast is maintained well at F1.8, and holds up on stopping down to about F4. Diagonal flare patterns become visible at F5.6, and shadow detail starts to deteriorate too. Both effects becomes more pronounced at smaller apertures, and at F16 we see well-defined and intensely-coloured flare patterns, along with substantial loss of shadow detail.
In the fourth sample we've zoomed in to 35mm and placed the sun just outside the frame. The lens does particularly well here at maintaining contrast and shadow detail.
|18mm, F1.8, sun in corner of frame||18mm, F5.6, sun in corner of frame|
|18mm, F16, sun in corner of frame||35mm, F5.6, sun just outside frame|
Background blur ('bokeh')
One genuinely desirable, but difficult to measure aspect of a lens's performance is the ability to deliver smoothly blurred out-of-focus regions when trying to isolate a subject from the background, generally when using a long focal length and large aperture. The 18-35mm gives you a decent degree of control over depth of field due to its large maximum aperture.
As always you'll get the most-blurred backgrounds shooting at the long end of the zoom, and here the 18-35mm's rendition of out-of-focus areas tends to be pretty attractive. The examples below give some idea of what to expect. In the first the subject distance is somewhere around 3m, which means that the degree of background blur isn't huge. But it's sufficient to focus attention on the fountain, and the blurring is generally attractive.
You'll get much more background blur when shooting at close focus distances, as always, and this is shown in the second example. Again the effect is smooth, although here the just-out-of-focus regions in the lower half of the frame show slightly harsh edges.
|35mm F1.8, Canon EOS 100D||35mm F1.8, Canon EOS 100D|
|Background detail, upper centre||Detail crop, upper centre|
Overall though, the lens's wideangle-to-normal focal length range means that it can't match a cheap 50mm F1.8 prime if getting strongly-blurred backgrounds is a top priority.