5 Years and Counting, Fujifilm Medium Format Autofocus Lacks Precision and Accuracy for Erratic Sharpness + Reader Comments
Related: Contrast and Phase Detect Autofocus, Dan B, Eastern Sierra, focusing, Fujifilm, Fujifilm GFX100S, Fujifilm GFX50S, Fujifilm medium format, Fujifilm mirrorless, Live View, medium format, Mt Whitney area, peak bagging, precision and accuracy, reader comment
I’ll be doing a lot more manual focus in the field with the Fujifilm GFX100S.
I was reviewing some older review pages and came across this page and others... way back in 2017 with the Fujifilm GFX50S (which lacks PDAF pixels), I wrote the following in Aperture Series: View to Mt Whitney From Alabama Hills:
When I shoot landscape, I want reliable, consistent focusing behavior. Fujifilm cannot do so! I should be able to 'nail' this scene and scenes like it with no particular effort. The results here are at f/8, where the image ought to be spectacularly sharp. The GFX50S with the 120/4 feels like working on quicksand—virtually all the outdoor work I did with it showed this kind of focus instability. If one cannot reliably focus with high precision, the AF system is worthless. — May 19, 2017
At the time I thought it might be the Fujifilm GF 120mm f/4 lens at issue. And perhaps it could/might be involved. But it was not alone in showing issues, and recent work proves that little if anything has changed.
Clearly not a PDAF issue, since the Fujifilm GFX50S lacked PDAF focusing pixels. So we get nasty striping artifacts from PDAF but no benefits for general photography. I say to Fujifilm: I will pay extra for a landscape-oriented high-res camera that is free of those nasty artifacts and can focus properly every time. And I will work with Fujifilm to get it right. All Fujifilm has to do is to contact me and get the project started.
Woe to the landscape photographer who trusts Fujifilm medium format autofocus to deliver accurate and consistent AF results. A system that is hit-and-miss guarantees inferior results, even if the hits are 80%, which my studies (below) prove is wildly optimistic. Add-in pronounced focus shift with some GF lenses along with field curvature on top of lack of AF precision and accuracy, and you are guaranteed to see puzzling sharpness results across the frame, or in patch areas, etc.
What you think was your goof and/or lens softness is often neither. Ya gotta get focus right.
You have no margin for focusing error with tiny 3.76 micron pixels. While f/8 can hide your mistakes a lot of the time, you will not have optimal results, or at the least, the peak zone of sharpness will not land where desired. If you are going to shoot a 100MP camera with tiny pixels, you had better get it right.
What I would like to see from Fujifilm is a firmware update that is a “slow but highly precise and accurate” focusing mode. A design emphasizing focusing speed is absurd for many types of photography!
Sampler:
Aperture Series: View to Mt Whitney From Alabama Hills
Fujifilm GFX100S: Field Examples of Autofocus Errors
Fujifilm GFX100S: Focus Precision and Accuracy with Fujifilm GF 35-70/4.5-5.6 @ 58mm
Fujifilm GFX100S: Focus Precision and Accuracy with Fujifilm GF 35-70/4.5-5.6 @ 58mm
Reader comments
Dan B writes:
I have the GFX 100S, have rented all the GF lenses, and own a couple. I found after many hundreds of test shots that regardless of the lens used that I need to AF a single scene 4 times to have high confidence of getting an optimally focused image (or very close to optimum). This goes for close subjects as well as for landscapes.
Manual focus can sometimes hit it right off the bat, but I can't do it reliably; by MF I mean enlarging the live view image in the EVF to max magnification, and (other than wide open shots) using stopped down focus until f4 or f5.6 in order to mitigate focus shift. I haven't experimented much with focus peaking. For now I just use the multiple AF shot method, except for panos, which is MF, and I have to take more than one set to be comfortable with getting a decently focused series.
I find this to be a R-PITA in the field and then having to comb through many shots at the computer.
I wonder if Fujifilm doesn't know about the problem, or knows about it but can't do anything because of some inherent imprecision in the lens hardware, or knows about it but doesn't care because most people haven't found it to be much of a problem (such as landscape oriented photogs who like to shoot at f/11 and 16; and of course portrait photogs who usually do many AF shots per pose anyway).
BTW, I just submitted a proactive renewal for a full access subscription. I find your detailed testing and analysis to be uniquely valuable.
DIGLLOYD: I think most users hit that AF-ON (or shutter) and just assume the camera got it right and then never know because stopping down hides it pretty well and/or because small errors can be blamed on any number of causes and/or vary slightly and it all looks pretty good and then too many also look at images on a Retina display and cannot see it anyway.
Focus peaking is problematic; IMO it is not only greatly inferior vs regular manual focus, it has negative value: it obscures the image so it cannot be seen, and it has too-coarse of a discriminating power.
I cannot be certain of the absolute best focus with manual focus (MF) because Fujifilm has the softest and crappiest-looking magnified Live View on the market. Still, I can do MF far more precisely than autofocus can, even vs small-spot AF at 100% magnified Live View. This is easily proven just by looking in magnified Live View: manually focus first, hit AF-on and watch the image go soft, all too often. I’ve seen that with every lens they make and with certain lenses, it can be reproduced at-will every time.
The focusing motors on Fujifilm medium format lenses are a dumpster fire compared to Sony’s technology; it feels like a decade of technological difference. Fujifilm really ought to re-issue all their lenses with focusing motors using current technology, not the landfill stuff they’re using. And redesign at least 1/3 of the lenses for less than ideal performance for the 100MP sensor, or at least solve the quality control problems (bad).
As for whether Fujifilm knows about it, it’s hard to believe they do not, but they sure would not not like to talk about it publicly! Maybe the focusing motors are part of the problem, but I am certain that the AF focusing algorithms are inherently flawed too as there is obviously no priority on accurate focus because the camera gives up very quickly.
