To make a long story short, we've re-shot our studio scene shots of the Sony a9 with the FE 85/1. 8 lens, and they're much sharper. We apologize for misleading any of our readers, but it's a long story - see below.
To jump to the images, just click the button, but we do encourage you to read the full text as well.
Just take me to the pictures!
The Long Story
You may have noted on the studio scene page of our Sony a9 review that we admitted to having quite a bit of difficulty focusing the camera with the new Sony 85mm F1. 8 lens in magnified live view. The maximum magnification (x9. 4) on the camera LCD made it very difficult to fine tune the 85/1. 8 precisely. Multiple AF-S attempts yielded shots varying slightly in sharpness, but this is not unique by any means to the a9: all cameras exhibit some tolerance when it comes to critical focus of a flat chart (which is why we always manually focus). The only way to check focus on each shot was to shoot tethered and check each shot magnified on a monitor. Of course, every time we thought we'd nailed focus, we'd try nudging the camera or focus ring just a bit to make sure we couldn't do any better, and then realize we'd fallen off a bit.
And so the search began again and again, with the quest for perfect focus ending up a bit of a fool's errand. We finally tuned focus to what we thought was reasonable (we look for maximum aliasing in the central Siemens stars, and color aliasing in the text), and shot our entire studio and dynamic range tests.
Subsequently, we got lots of complaints about the a9 being soft.
The Lens Factor
Was it the lens? This is the first Sony FE camera we've shot without the stellar Zeiss 55mm F1. 8. We've had a long-standing policy of shooting with an on-brand 85mm equivalent lens per-system, to maintain equal distance from camera to target, something that allows for all images to be rendered with equal perspective. With Sony's recent release of the razor sharp FE 85/1. 8, we thought we'd stick to our policy and give it a try.
But we don't blindly switch lenses for a system; we first verify:
The new lens is at least as sharp as the previous one.
The lens transmission (also accounting for the aperture at F5. 6) is not so different as to affect noise comparisons.
Our initial testing showed equivalent sharpness between the 55 and 85mm F1. 8 lenses on even a high-resolution a7R II (see below). Furthermore, DXO verified similar levels of sharpness between the 85 and 55 F1. 8 lenses (which both perform better than Sony's 85/1. 4 GM, surprisingly). And while we don't have a way of directly measuring lens transmission, we measured signal:noise ratio of a few grey patches in our scene with the two lenses on the same camera body, and found them to be within 1/6 to 1/10 EV of one another. That meant the new lens would not make the a9 look better, or worse, in Raw noise comparisons compared to if we were to use the Zeiss 55mm F1. 8 at F5. 6.
Sony 85mm F1. 8 at F5. 6 (left) vs. Sony 55mm F1. 8 at F5. 6 (right). Shot on a7R II
Some Friendly Help
While plowing ahead with other aspects of the review, a message from forum expert Jack Hogan turned up in my inbox showing this:
Long-time forum member and all-round expert Jack Hogan did a quick MTF analysis per color channel based off of the slanted edges in our scene. Uh-oh. Looks like the red channel is focused better than the green channel, yielding a calculated MTF50 of only 945 line pairs per picture height.
Importantly, the green channel should have the highest MTF.
It was now clear that focus was the underlying issue with our studio shots. Not a bad lens. Not a strong anti-aliasing filter. But simply the fact that the lens was not optimally focused: if it were, the green channel would have the highest MTF.
So we sat down one day and spent the entire day shooting many, many runs of our studio scene, slowly moving a macro rail (rather than coarsely adjust focus on the lens) between each run. From these shots, we picked the (centrally) sharpest runs. While our copy of the 85/1. 8 appears slightly decentered (the left is softer than the right), the results now are much more in line with where things should be:
Jack Hogan re-analyzed some of our new studio shots of the a9, and the green and blue channels now have the highest MTF, not the red channel. The calculated MTF50 of 1125 lp/ph, which is a 19% increase in linear resolution over our previous results.
A side benefit of analyzing properly focused shots is an ability to estimate the strength of the anti-aliasing filter, which appears to kick in around 0. 744 cycles per pixel (the first minimum in the MTF curve). For comparison, the D5's anti-aliasing filter kicks in around 0. 748 cycles per pixel according to Jack's analysis of our studio scene shots. Meaning the a9's AA filter is fairly typical.
Have a look at our updated images, and our updated image quality analysis based off of our new results:
See our updated studio scene
Editor's note:
As camera sensor and lens resolutions are becoming astronomically high, tiny little differences become visible in pixel-peeping. And that's precisely what our studio scene allows you to do. Remember that even within the depth-of-field, there is a plane of peak sharpness.
Our studio scene isn't perfect, but it can be helpful. It has its caveats though. For example, because we don't control for lens transmission from brand-to-brand, or any shutter speed inaccuracies, we state that noise comparisons are only accurate to within 1/3 EV. Trying to extrapolate differences smaller than that from high ISO shots of our studio scene is meaningless: margins of error are real.
The same goes for sharpness. The reality of lenses and mounts is that there is copy variation - in both. Therefore, we urge you to make sharpness comparisons largely from the center of the scene, which removes the lens (as much as it can anyway) from the equation. The rest of the scene is useful for assessing color, detail retention and noise at high ISO in JPEG and Raw, respectively, and other subjective attributes. And keep in mind common sense things: the lock of hair is well above the plane of optimal focus, and different lenses can have field curvature which either helps or hurts the sharpness of this lock. It's important to keep these sorts of things in mind when pixel-peeping our scene.
This time, with the a9, we take full responsibility for a non-optimally-focused set of shots. But the process has also been a learning experience for us: depending on a lens' electromechanical coupling and the magnification of the live feed, it can be extremely difficult to take test shots that stand-up to the level of scrutiny our image comparison tool demands. And there are the practical issues mentioned above around taking one shot, checking it, and repeating the process - returning to the position of optimal focus is nearly impossible. The results of visually checking which shot is sharpest can even vary from tester to tester. I can assure you though: we are constantly working on methods to improve these processes.
That said, it's important to keep things in perspective: in the real world it's unlikely you'd have seen the sharpness 'issues' we had with our initial a9 run (that otherwise appeared so drastic in our studio scene). Why? Because (1) you don't typically view images at 100%, (2) there will at least be a plane of maximum sharpness (which in our case, unfortunately wasn't our studio scene on our first run), and (3) your lens and shooting aperture will have far more impact on subject sharpness than which 24 MP sensor was used to shoot it.
To our readers: we offer our sincere apologies, and wish you happy shooting!
A special thanks to forum member Jack Hogan.
. dpreview.com2017-6-23 19:03
