This is the second lens in Panasonic's recently announced, premium Leica DG Vario-Elmarit F2.8-4.0 Series. All these lenses are expected to deliver superb performance, at the same time, thanks to variable aperture, staying within reasonable limits size- and weight-wise.

This is not the first super-wide zoom for Micro Four Thirds. As of this writing (June, 2017) there are three such lenses on the market already, plus one in the older Four Thirds mount:

For a rectilinear lens, the horizontal image angle is α = 2×atan(d/2F), where F is the focal length, and d — the frame width (17.3 mm). To get the diagonal angle, use 21.6 mm instead.

The FT lens was historically important. It was released in 2005 and for a good few years it remained the widest rectilinear lens (zoom or fixed) for any frame format smaller than the "full" 24×36 mm.

First impressions

The lens is not as heavy or as large as the FT and μFT 7-14 mm ones from Olympus. It is similar to the MZD 75/1.8, in shape, finish and feel, but not in size, being about 2 cm longer.

Zooming is mechanical, manual focusing — electronic. Both rings are exemplary in their balance between smoothness and friction. The zoom ring has markings at 8, 9, 10, 12, 14, and 18 mm.

The front elements move in or out during zooming action. The lens barrel, however, sits inside a protective tube of fixed length, so the lens movement remains hidden. It is that tube that houses the filter thread (67 mm) as well as hood bayonet.

As a result, the distance between the lens and a mounted filter depends on the focal length and may change within a centimeter or so.

The lens hood itself is of a premium type, with a nice unlock button and the right amount of spring action where needed.

Lens feels good on the E-M1 II, but even better on Mk.I. No problems on E-M5 and, surprise, quite nice on E-PM2. You just have to grab the whole thing by the lens, not body.

Specifications — and why

Some of these are trivial, some not. Let me walk through all, just in case.

• Focal length: 8-18 mm.

  This is equivalent to (i.e., provides the same field of view as) a 16-36 mm lens mounted on a "full frame" camera. With 7 mm being the ultimate focal length for a μFT rectilinear wide-angle lens, this lens comes a close second. The extra 4 mm at the long end may help in reducing frequency of lens changes when used along a "standard" zoom starting at 12 or 14 mm.

• Maximum aperture: from F/2.8 at the wide end to F/4.0 at the long one.

  Focal length	8 mm	9 mm	10 mm	12 mm	14 mm	18 mm
  Maximum aperture	F/2.8	F/3.1	F/3.2	F/3.4	F/3.6	F/4.0

• Minimum aperture: F/22.

  Using F-numbers of F/11 and beyond is an option only if the depth of field is more important than the negative effects of diffraction. Usually you will want to use F-numbers of F/5.6 and lower.

• Minimum focus distance: 23 cm from the image plane at all focal lengths.

  This is eleven centimeters from the lens filter thread, quite close.

• Maximum magnification: 0.12× at F=18 mm.

  This is the same as 0.24× on a "full frame" camera, and it corresponds to a field of view about 11 cm (4.4") across.

• Filter size: 67 mm.

  More often than not in this type of design, the front element, while being concave as a whole (thinner in the middle) has a strongly convex front surface (protruding center). This usually precludes, or at least complicates, filter usage.

  Not in this lens, though. Here we can use filters, and two kinds come to mind: protective (or skylight), just to keep the front surface clean, and neutral-density, to allow outdoors shooting at full aperture — to get at least some control over the depth of field. (See my bobble-head pictures above full-size.)

  With a filter mount 4.5 mm thick, I could not find any traces of corner cut-off, even when stepped down all the way to F/22.

• Dimensions: 73 mm diameter, 88 mm length.

• Weight: 315 g (11 oz).

  That's about 70 g less than the weight of the MZD 12-40/2.8, not bad at all.

• Drip/dust-proofing: yes.

  Nice to have, but adds to size, weight, and price. I realize mine is a minority voice on this subject.

In the field

The autofocus action is fast and accurate, zooming smooth and precise, so seems manual focusing assisted by peaking (although I don't see a good reason to use MF here). The lens handles well and has this pleasant feel. As expected.

Being able to go all the way to the "long wide angle" of 18 mm is especially nice, as I can shoot in almost-normal perspective without switching lenses. This turns out to be more useful than I thought.

After all, the classic definition of "normal" lens assumes focal length equal to the frame diagonal (here: 23.6 mm), and 18 mm is 83% of that value, fairly close.

The lens does not have its own image stabilization system. It will, of course, benefit from Olympus in-body IS. I expect to have a fighting chance for decent one-second hand-held exposures at 8 mm.

Let me pause writing for a moment, grab the camera, and squeeze out three frames: 1-second exposures, handheld, sitting, elbows and back unsupported. And then another three for a different focal length. Yes, 50% success rate.

Performance: the MTF data

I found two MTF graphs for this lens, published by Panasonic: full aperture at both extreme focal lengths. If you are not familiar with MTF graphs, then perhaps have a look at my small article on this subject. I am also explaining there, why I consider the MTF graphs in this form not very useful for lens performance description.

In similar circumstances, blue (contrast) lines stay above 90% up to x=6 mm at least, for a good lens. This is easily exceeded here for both focal lengths, an excellent result.

For comparison, here are similar graphs for the MZD 7-14/2.8 at 7 mm and 14 mm.

For the green (sharpness) lines, things are less clear. Many manufactures, including Olympus, use here a frequency of 30 LP/mm; Panasonic — 20 LP/mm (that's what their "40" means), which makes the data directly not comparable.

Still, it is often assumed that for 30 LP/mm the strong performance region starts at 80%, and for 20 LP/mm — below 90%. As the Leica lens meets both these requirements, we can expect from it a very solid show.

Note that the lines start to drop off quite far to the right, above 8 mm or more. 8.65 mm is half-width of the frame. This means that loss of resolution/contrast really starts at that boundary, also a good sign.

Actually, based on MTF data alone I would expect the Leica 8-18/2.8-4.0 lens to be better than MZD 7-14/2.8, especially off-center.

Vignetting

The light fall-off (or vignetting) is an effect seen mostly with wide-angle lenses, especially at full apertures. It shows itself as gradual darkening of image areas away from the center. Some of it is due to the lens itself, some to the directional response of the sensor, and some — just to geometry. Lens and camera engineers may try to reduce the magnitude of the first two parts, but they can do nothing about the third one (except for moving the lens away from the sensor).

At an angle of 33°, light becomes only half as effective as at zero; at 45° — four times less effective.

A correction for vignetting is available as an option on most cameras. In Olympus ones it has to be explicitly enabled, found in the menu tree as "Shading Compensation".

Various implementations may differ in when (at which processing stage) the correction is applied. This may range from the raw photosite response to the full RGB pixel information, but the general principle remains the same: amplifying the response at each point, depending on where in the frame this point is (of course, for many points the multiplier will be 1, so those are left alone). And if you amplify the signal, you also amplify the noise in that area. This is the price we pay.

Interestingly, in Olympus cameras of the last ten years or so, the correction seems to be applied during analog-to-digital conversion of the photosite signal (not even at the raw-to-RGB stage), so that the raw files are affected.

This is indicated by the fact that the Auto Shading Compensation option in Olympus Viewer is disabled for raw files produced with the in-camera correction active.

The most vignetting you are going to see with this lens will be at 8 mm, between F/2.8 and F/3.5 or so (there is no improvement going from F/4.0 to F/5.6). Have a look at these two reduced-size images (or, if you wish, full frames).

Looking closely at these samples (by flipping them in an image browser; otherwise it is hard to see anything), I can see that the vignetting here is not really heavy (dark), but it is quite extensive, taking a large share of the image area. This is different than on many lenses I've seen, where the effect starts only at distances beyond frame half-width, rising fast towards corners.

Quite interesting: the metering system exposed both frames identically, basing its decision on the central 1/3×1/3 (or 10%) of the image area, where vignetting does not reach.

More, a subtle vignetting effect may enhance a photograph, adding an extra accent on the central subject area and de-emphasizing the peripheries. I really like the effect and would rather keep it, for some applications at least.

At 12 mm the full aperture is F/3.4. There is less vignetting, but it is still of the same gently-sloped type. All is gone just 1/3 EV down the road, at F/4.0.

No discernible vignetting at 18 mm.

The results with in-camera correction enabled were as expected: no vignetting, except at 8 mm, F/2.8, where about half of the effect went away.

Geometric Distortion

Wide-angle lenses generally suffer from a problem of image magnification changing with the distance from the axis. This leads to barrel- or cushion-like distortion, especially visible in long lines, parallel and near to the image edges. The shorter the focal length, the more difficult is this effect to avoid.

In fisheye lenses, designers give up on trying to correct the distortion, focusing on other optical flaws instead. This lens, however, is rectilinear, supposed to keep straight lines straight.

Distortion, however, is (like vignetting and, partly, chromatic aberration) not so difficult to remove after the exposure: for a given, known lens its behavior is predictable. If some point of the subject is rendered at a point P of the image, we can calculate the point Q at which it should be rendered so that the image is not distorted. Then we can move the RGB information from the pixel at P to the pixel at Q; if we do that for the whole image, the result will be distortion-free.

I'm simplifying things here — but not by much. Moving the pixel RGB information from P to Q can be replaced with interpolation or combined with demosaicing of the RGB matrix, but the basic principle stays the same.

Olympus and Panasonic chose to go this way. Some of their lenses provide only a most rudimentary degree of correction for geometric distortion, and most of work is done in image processing.

There are three degrees of distortion correction in μFT lens/camera combinations:

• Optical — no correction (except that by the lens optics); each pixel stays in place. You can't get this version with Olympus (or Panasonic) software; you have to convert a raw image to RGB using a third-party program.
• Basic — the correction as described above is done in-camera during the raw-to-RGB conversion. For raw files, the same correction is applied during such conversion in Olympus Viewer. It is not optional, there is no way to disable it.
• Post — additional correction, applied optionally at the postprocessing stage in Olympus Viewer or other software. The Viewer has an option to do it hands-off, using the lens data passed (or identified) in the raw file.

The Distortion Experiment

Let us see how this works for our lens. I've shot two series of pictures, each for a scene including a straight line, going through the whole frame width, close and parallel to its top. For each picture, I read three distances, all measured vertically from the frame top:

• a₁, a₂ — to the leftmost or rightmost end of that line;
• b — to the midpoint of that line.

We also need to know the value of h, half of frame height; for μFT (20 MP) it equals 1944 pixels.

The so-called "new" SMIA TV Distortion can be then computed as

I was saving both raw and JPEG files, so for every shot (focal length) I could generate images in all three correction flavors and compute the distortion as shown above.

Scene 1: Roof Line

This was shot from a distance of about three meters; four focal lengths were covered.

The first numbers I've got were the ones for in-camera JPEGs. They looked so typical for a high-quality super-wide zoom, that I was going just to make a statement to that effect an close the matter. Only then I remembered the controversy about fixing lens flaws in software. Time to start digging!

Running the raw files through Olympus Viewer confirmed my understanding of the Basic and Post distortion correction and demonstrated that the "hidden" part used in-camera is identical to that in Viewer.

The table shows complete results for that session. The line for uncorrected results is quite shocking. If you would like to know what eight-percent distortion means, have a look at the pictures below, where the Optical version with -8% is compared against Basic with -1.6% (done in-camera).

And that's the controversy I was talking about. Should the lens quality be judged by the first picture, or by the second one? Should we consider this cheating? We will come back to this later.

Note another effect: some areas of the original, captured image are missing in the final version (the yellow container, chair at far left). Obviously, correcting for distortion affects the effective angle of view, or the effective focal length.

Scene 2: TV Screen

The subject of my second scene is a large-screen TV. Six focal lengths were covered this time. In order for the top edge of the screen to fill the frame width, I had to get closer: this time about 50 cm from the subject.

Distortion depends on the subject distance, and the numbers seem to confirm that. Still, the general pattern remains the same: huge distortion at wider angles without software correction, greatly cleaned up with Basic processing.

The effects of Post correction at a close distance don't seem to make much sense. No big loss: I'm happy with the Basic in-camera results.

Which distortion?

So, which one of these three values should I use to judge the performance of this lens with regard to geometric distortion? As a systems engineer, I would opt for the second one (Basic), treating the lens+camera combination as a black box with some kind of standard, no-hassle output. Therefore I stick to my middle numbers shown in the tables above.

Why don't the designers opt for a complete removal of distortion in one step? I think the reason is the computational complexity. The Basic approximation calculations, applied in-camera, have to be fast. Whatever is not fast enough must be left for postprocessing.

Interestingly, the viewfinder image seems to be largely free of distortion. This means some correction is applied to it in real time, 120 times a second! (Some tricks could be used to cut this frequency down, but it still has to be fast.)

I wonder whether this is the full Basic correction, or another level, simpler and faster. I consider the former quite feasible: after all, they already do raw-to-RGB conversion real-time, and most of the correction algorithm is (the way I understand it) just substitution of pixel index values with shifted ones.

Because most of distortion is now handled in the software (and so is most of chromatic aberration, a very similar story), the lens may be now designed to address some of the other aberrations more efficiently.

EVF correction

In the last section I wasn't sure how much of distortion is corrected in the viewed image, real-time. Is this the full Basic correction, or something quick-and-dirty, just to keep up with the finder refresh rate?