Sigma 12-24mm f/4 HSM DG ART ( Canon ) - Review / Test - Analysis
Lens Reviews - Canon EOS (Full Format)

Distortion

Ultra-wide zoom lenses tend to have a rather extreme distortion characteristic and that's also true for the Sigma 12-24mm f/4 HSM DG ART. At 12mm, the barrel distortion exceeds 5%. This will be easily noticeable whenever you shoot architecture and the likes. Of course, the issue eases when zooming out. At 14mm, the barrel distortion is comparatively moderate at 2.4% and beyond it's not really something to worry about.

Vignetting

The vignetting came as a positive surprise. It's far from being undetectable at the wide end, but a light falloff of "just" 2.2EV (f-stops) at 12mm f/4 is very moderate indeed for such a focal length. Sigma probably had to over-design the lens a bit for that. Stopping down to f/5.6 reduces that to 1.4EV and it's very acceptable (for a full format lens) from f/8 onward. At the other tested focal lengths, you may still spot some vignetting at f/4 but it's sufficient to stop down to f/5.6 in order to tame that.

MTF (resolution) - at 50 megapixels

The resolution characteristic of the Sigma lens is very mixed. It has two souls here really. The image quality is surprisingly good at the difficult 12mm setting. The center quality is already great at f/4. The outer image field is a bit soft at fully open aperture but recovers substantially at f/5.6. Stopping down to f/8 and f/11 reduces the technical quality a bit but the borders/corners remain good to very good. f/16 should be avoided due to diffraction. The overall characteristic is similar at 14mm. At 18mm there's the first drop in performance which is most apparent at f/4 where the corners aren't pretty anymore. Stopping down to f/5.6 boosts the quality to "normal" levels. The problem maker is the "far" long end. The performance collapses here - not so much in the image center but the corners are rather unusable at f/4 and still soft at f/5.6. Sticking to f/8, better f/11 is a good idea here also because the field curvature (curved focus field) is massive at this setting - see also the next chapter.

The centering quality of the tested sample could have been better really.

Please note that the MTF results are not directly comparable across the different systems!

Below is a simplified summary of the formal findings. The chart shows line widths per picture height (LW/PH) which can be taken as a measure for sharpness. If you want to know more about the MTF50 figures you may check out the corresponding Imatest Explanations

MTF (resolution) - at 21 megapixels

Why are the MTFs sometimes "better" on 21 megapixels compared to 50 megapixels ? There are two reasons for this. Lateral CAs are lower in terms of pixel widths at 21mp simply because the pixel density is also lower. Extreme CAs that may exist at 50mp are therefore less affecting the measurements at 21mp. Generally we are also using a certain degree of sharpening during the image conversion (just like in real life images) and because the 21mp results are "sharper" on pixel level they are relatively more receptive to (mild base-) sharpening.

Since most users are still using 21mp DSLRs, let's have another look how it performs in this scope then. Of course, the lower resolution is boosting the charts. At 12mm and 14mm the performance is very good across the image field at all mainstream aperture settings. At 18mm, there's drop at f/4 with some softness in the image corners but this is recovered at f/5.6. The "long" end of the range remains, of course, the weak spot. Stopping down the f/8 (better f/11 due to field curvature) is advisable.

Please note that the MTF results are not directly comparable across the different systems!

Below is a simplified summary of the formal findings. The chart shows line widths per picture height (LW/PH) which can be taken as a measure for sharpness. If you want to know more about the MTF50 figures you may check out the corresponding Imatest Explanations

Field Curvature - at 50 megapixels

We always feel a bit uneasy when providing graphs about field curvature. So please note our disclaimer that the following isn't an overly accurate way of showing the effect.

In the MTF chapter we showed you the resolution on the focus field. In a perfect world scenario the focus field is absolutely flat - so e.g. if you take a perpendicular picture of a wall, the focus field should sit exactly on that wall. Now that's the theory - in practice this is almost never the case really. Now normally that doesn't really matter. First of all you rarely take pictures of a wall anyway - usually these are not overly exciting - and the depth-of-field is often hiding these imperfections to some degree anyway.

That being said some lenses are better than others here. And the Sigma is not a shining example how it should be. The focus field is quite flat at the wide end of the zoom range. However, things go downhill from 18mm onward and the situation is quite bad at 24mm. Again, take the following with a grain (or two) of salt - below is the MTF graph at 24mm taken on the flat plane (thus using only marginal focus variations from the image center). It's not pretty as you can see. The far corner quality is pushed into non-existence and the borders aren't usable either till f/8.

In case you wonder about the "worse" border values at f/5.6 vs f/4 - this of a curved focus field as the waves from a water drop. The waves have different amplitudes over time - and the focus field has different hot spots depending on the aperture.

Updates: We have chosen a reference setting that ignores the extremes of the field curvature effect - which is more realistic.

Field Curvature - at 21 megapixels

Let's also revisit the field curvature at 21mp now. Of course, the situation has relaxed here but f/4 and f/5.6 should still be avoided. The depth-of-field catches up at f/8 and the field curvature effect is much reduced at f/11. Please note that this is the performance on the flat focus field.

Chromatic Aberrations (CAs)

Lateral CAs are surprisingly low for such a lens. Generally they are smaller than 1px on the average at the image borders. The only exception is the 24mm setting at f/4 where the CAs peak at 1.5px but even so this is very moderate on a 50mp sensor.