Petapixel has a recent opinion piece which boils down to the claim that DSLRs have faster AF than mirrorless cameras, especially in low light, because they have a “dedicated autofocus module, separate from the main imaging sensor”.
Is the claim (faster AF in low light for DSLRs) actually true? I thought recent (>2020), mid- and top-tier MILCs were at least as good or better than DSLRs of the same class, for the same manufacturer (eg Nikon Zwhatever to Dwhatever).
My understanding is that the PDAF “pixels” on the sensor have a separate readout path, and the limiting factor these days is usually lens movement.
Since I have access to both systems currently, here’s what I found out. I tested Z5 ii vs D3500 in a dark bathroom without windows. Important to note that Z5 ii has hybrid AF and the camera mostly decides automatically when to use PDAF or CDAF.
With an F/4 lens
the D3500 really struggles with PDAF (it rarely locks focus, if ever) and the CDAF is completely hopeless
Z5 ii takes its time sometimes, but it’s consistently successful and it impresses me what it can lock onto
With an F/2.8 lens
the D3500 starts to catch up with PDAF, it’s a lot easier to lock focus, but CDAF is still hopeless.
I don’t have a 2.8 lens for Z5 ii, unfortunately.
It doesn’t exactly answer the speed factor, but success rate is much better with mirrorless. Speed seems to be lens dependent as those AF-P lenses are really fast to focus, even compared to the Z5 ii’s kit lens
EDIT:
I will also say that the conditions were kinda extreme, the Z5 ii metered about 1/13, f/4 and ISO 51200, so yeah. The objects were high-contrast, but lit only with a dim night light. When the light slightly increases, the DSLR starts to pick up nicely, but I almost can’t see **** in the OVF
Hi, I agree with the comments that say that he does not mention which cameras and lenses is comparing.
It is an opinion article without data, for example, there are comments giving objective data(not checked from my side) but the best DSLRs are able to focus at -5EV and some mirrorless -8EV or -10EV(depending camera and lens).
I’m also a bit curious about that and I don’t think the spec sheet tells much in this case.
I’ve got my hand on a Nikon D5200 some times ago which I believe is rated to focus down to -1 ev, and I also had a Pentax K-70 which is also rated to focus down to -1 ev for the 2 linear type sensor at the two sides. I tested with Nikon’s 18-105mm and Pentax’s 18-135mm both at the long end at F5.6, the Nikon outright refused to focus at dusk when there is still some light in the sky, which shocked me because the Pentax will just focus in most cases in my previous experience with it (though in very dark light, the in-focus rate is close to 50-50, so I’d use the -3 ev rated cross type points in such scenario).
So, my take is, two things that look the same on spec sheet might be different in real life, and I’m hesitate to reach any conclusion based on the spec sheet provided by the manufactures now.
This was probably true around the time both systems were relevant (2018 - 2020?). It is very difficult to test this today as the most modern high performance DSLR cameras have been introduced in 2017 while new EVIL* with highest specs drop every week.
Comparing a D3500 (lowest end APS-C) with a recent full frame body isn’t really fair.
I never thought the autofocus on my D850 was exceptionally fast. I think the same for the Z7ii. And I don’t generally photograph things that move, so this is still on a tripod, making a composition.
My D5100/D5500 front- or back-focused quite often, I kept adjusting with the hex screw near the mirror (since these models don’t have micro-adjustment), but eventually gave up because different lenses would need to be adjusted differently.
Switched to the X-T20 and X-T2, focus is always dead-on, even with the 56mm/1.2. Now that I think about it, I do shoot at indoor events (theater, recital, dance, karate tournament, …) and rarely need to have faster AF.
Frankly, I find the current discussion about AF ridiculous. It’s all “subject detection AI tracking 3D” autofocus. As if everybody required split-second accurate tracking of birds in flight.
In another forum one member seriously complained that a camera did not get a 90% hit rate on a hummingbird’s eyes at 120 FPS. I am not making this up.
Admittedly, the most demanding thing I shoot is pictures of my kids. They run and jump and scream and fight. But two or three frames per second is more than sufficient. Focusing on their faces has been a solved problem for a long time. To say nothing of landscapes, of course.
This is my number one reason for switching to mirrorless - this problem simply doesn’t exist in the MILC world. I shot with 18-55mm DX AF-P for about four years till I realized that it front focused from 35-55mm and I wondered why my shots are not sharp the same for the Sigma 17-50mm f/2.8 - very bad front and back focus at the same time, AF fine-tune was basically useless. That’s when I definitely decided for mirrorless that I used to look down on… and now I’m in
I think that DPreview was largely responsible for generating this sentiment, every review starting from around 2020 has these kind of tests. For a few years they always just concluded that Sony is the best and you just cannot beat it, period, but other camera makers have closed the gap since.
It took me about 2 years to learn how to almost always get a shot in focus with a contrast-based camera AF on a camera that can shoot at least 5 frames/s (no refocus). Knowing the DoF of your lenses at various apertures is key. Focus bracketing is also useful.
With a digital camera, you have a near-zero marginal cost per shot so you can almost always get a shot. The difference between a 90% and a 99% “hit rate” is 10% more shots… basically an unimportant detail.
It’s kind of impossible to test because equivalent tech isn’t available. Even if you consider DSLR af frozen with the last action focussed DSLR the milc lenses will be way more recent or high end.
Would be interesting to hear someone with expertise int the actual tech make an opinion on the theoretical winner. No chance of that happening though as said experts will be paid to sell milc. Users are a bit the same and have preferences either because they’ve invested in pricey new gear or are nostalgic for an older experience.
Its generally impossible to detangle things now because DSLR vs milc comparisons were never even. When lenses were compared the milc lenses used software corrections that were of course equally available to DSLR lenses but just not auto applied in lightroom. AF test with milc for the longest time only discussed the animation in the viewfinder not actual hit rate.
“Does it detect eyes at 30m distance? No sadly only faces, inferior product” despite dof at said distance being such that eye detect is meaningless. There was just so much nonsense going on because the tech was new.
I use both the Nikon D500 and the Nikon Z8, the former one of the best DSLRs ever made and the latter certainly one of the best current mirrorless camera as far as high-end AF is concerned. Both for bird photography. Gotta say, the Petapixel article is just wrong, at least in this case. The Z8 works much better in low light.
I will grant that one thing the D500 does a little better is rack front-to-back because that’s the way its algorithm works, but that’s not really a criticism of mirrorless and they have to be used slightly differently. And then you have to add that the D500 (and every DSLR) is less accurate than mirrorless. Overall, the Z8 works better, period.
It doesn’t make sense to say that DSLRs work better because of separate processors. Rather irrelevant. What matters is data processing speed and the algorithms. Two separate processors can be weaker than one single one if the latter is faster than the two separate ones combined. For the same reason, just Ghz and number of cores can’t tell you alone whether one computer is faster than another. You’ve got to consider the other optimizations like SSE, memory, architecture, etc.
But especially for low-light shooting, mirrorless typically wins for speed and accuracy these days.
Not only that, there are several years of processing power and software algorithms advancements that skew the data in favor of newer cameras.
Regarding the “user opinion”: this is the long term review of a professional wildlife photographer regarding Canon’s R1:
Basically he says that the R1’s autofocus is the best system he has ever used, period (which says a lot coming from somebody used to the 1DX). But crucially, all the things he describes are software, not hardware (“amazing eye tracking”, “problems with falling snow”). Canon has been using PDAF on the 1D series from several years back, so that part should be well tuned. But the R1 has an “accelerator” processor, that’s capable of running ML and algorithms really fast.
You are not going to believe me but I have been trying to get my camera (D7100) to check one (kinda funny) thing… set a low ISO, close the aperture and fast shutter speed to achieve -5EV, well the OVF only shows 2-3 EV steps but I’m almost sure that I will see through the OVF hehe
the same for the Sigma 17-50mm f/2.8 - very bad front and back focus at the same time, AF fine-tune was basically useless. That’s when I definitely decided for mirrorless that I used to look down on… and now I’m in