Viewfinder AF sharper than Live View?

[Matthew Currie]

<p>I dropped my D3200 a while ago, with the 55-300 lens on it, and although it seems to be all right, I thought I'd do some controlled testing to make sure the AF is OK and the lens reasonably sharp.</p>

<p>So I set it up on a tripod in low available light, and took a few pictures of things with the lens set at 300, the aperture at F8, first with viewfinder AF and then with Live View. I generally use back button focus, AFC and dynamic area focusing. In Live View it's at normal and single servo. I did it first with the self timer, and again with the infrared remote. Now I obviously don't have much to complain about in the calibration of the camera, because in repeated tests, viewfinder autofocus was visibly better than Live View except in one instance when it was indistinguishable. Given that this is not as it usually as expected, I am wondering if anyone has run across this before. </p>

<p>In the process, I discovered another in a growing series of undocumented anomalies in the operation of the D3200, too. When back button focus is enabled, AF operates as you'd expect in viewfinder mode, and in Live View with the shutter button and the self timer, but the remote overrides the back button and triggers AF. </p>

[5711]

<p>viewfinder af is called phase detection.<br>

live view works by focusing over contrast.<br>

read up on this.</p>

<p>f8 with view finder af in low light...thats just wrong.</p>

<p>to get you started, id recomment you read this blog:<br>

http://www.wildlifeinpixels.net/blog/</p>

<p> </p>

[5711]

<p>viewfinder af is called phase detection.<br>

live view works by focusing over contrast.<br>

read up on this.</p>

<p>f8 with view finder af in low light...thats just wrong.</p>

<p>to get you started, id recomment you read this blog:<br>

http://www.wildlifeinpixels.net/blog/</p>

<p> </p>

[mag_miksch]

<p>phase detection is much quicker and more reliable at bad light but it shouldnt focus "sharper", "in focus" is "in focus"</p>

[Matthew Currie]

<p>By "focusing sharper" I obviously mean what the camera accepts as correct focus. I have read up on it, which is why I'm posting here and braving the inevitable condescension, because everything I've read suggests it should be the other way around. </p>

<p>I don't see why F8 with viewfinder AF in low light should be "just wrong" for an experiment in which, clearly, the exact content is irrelevant and the camera is on a tripod. As far as I know, AF occurs at open aperture, which in this lens at 300 is 5.6. If it focuses at all (which it did) then it's focusing. When I switch to normal focusing, this camera operates only in focus priority, so if it fires, it has achieved what the camera considers correct focus. </p>

<p>I have tried this experiment in daylight, with the kit 18-55 lens as well, at various focal lengths and apertures on various subjects. I tried resetting the AF from C to S, in case that made a difference, and tried it with AV on and off, in case the supposed tripod sensing of the lens was off, using both back button and normal focus. In cases where a difference is detectable at all, viewfinder focus is almost always sharper than Live Focus. 100 percent views at the telephoto end are most obvious. A little difference can be seen with the shorter lens at close range. As expected, it disappears at wider angles where focus accuracy is less critical. </p>

<p>I remain a little mystified, though hardly disappointed. It appears that the camera's auto focus brain is a little more precise in viewfinder view than in Live View. </p>

[chuck]

<p>I don't believe view finder (phase detection) AF is sharper than LV. I have personal experience to prove it if you would read the my most recent thread.<br /> But beyond that, I think there are strong theoretical reasons to suppose view finder AF will not be sharper or less error prone than LV. Basically, they boil down to the following:<br /> 1. LV AF is immune to minor misalignment of camera parts. LV AF works by reading the actual image as seen by the image capturing sensor. If the LV image is sharp, then actual picture taken, which is the very same image, will be just as sharp. Phase detection AF, on the other hand, uses a separate image reflected into the sensor by a system of lenses and mirrors behind the main mirror. The key is phase detection sensor is not reading the exact image the image capturing sensor is reading. In theory the phase detection AF should be placed at exactly the same distance away from the lens as the main image capturing sensor so when the image at one is in focus the image at the other is in focus as well. But it is certainly possible for some misalignment to make that not to be the case. <br /> 2. LV AF is immune to aperture effect. It is actually reading the two dimensional image on the image capturing sensor. So doesn't care about how big the aperture was that the light had passed through on the way to the sensor. It will work at F22 as well as at F1.4. Phase detection sensor, on the other hand, is concerned not just with the actual image on the AF sensor, it also relies on information about the angles of paths the light took getting there. So the wider the aperture, the bigger the range of angles of the light falling on the AF sensor, and better the AF sensor works. So AF sensor will work better with a F/2.8 lens, then with an F/5.6 lens.<br /> The reason why companies like Nikon and Canon, who build much of their reputations on big lenses used to capture rapid actions at sporting events, continue to use phase detection AF is because phase detection AF, by virtue of relying on the angle of the path of light in addition to the image on the sensor, can tell how far out of focus an image is, and in which direction and by how much the lens has to adjust in order to bring the image into focus. So a single reading from phase detection AF is sufficient not only for telling the camera whether it is in focus, it is also sufficient for telling the camera in what direction and by how much it should adjust the focus to being the image into focus. So phase detection AF is much faster to achieve focus, and much more capable in tracking a moving subject whose focus is continuously changing.<br /> LV focus can't do that. It knows when the image is in focus, and when it isn't. But when it isn't, LV focus can't tell how far from being in-focus it is. It also can't tell in what direction it must adjust the focus to bring the image into focus. It has to simply turn the lens to try to look for focus. So it doesn't do very well trying to anticipate subject motion and rapidly tracking the motion.<br /> So this probably explains why Nikon and Canon is sticking with single lens reflex mirror design with phase detection AF, where as Sony isn't. Sony has negligible presence in the action photography market, and is trying to distinguish itself by offering the best sensor for static images. So Sony, unlike Nikon and Canon, has little need of fast action tracking and predict focus. Sony only need supreme accuracy in focus, however long it takes. Hence Sony ditched the mirror and the a real phase detection AF sensor, and relies on its equivalent of LV AF. Nikon and Canon needs to be able to track motion and predict focus, while they also would like to have the most accurate focus when speed is not of the essence. So they kept the mirror, the phase detection AF and the also have LV AF.</p>

<p> </p>

[Matthew Currie]

<p>It's hard to make a visual comparison here, owing to the size restrictions and the loss to JPG conversion, but here is a small example, cropped from Raw images in View NX2 and converted, then placed side by side. The resulting 781 pixel wide image was resampled to 700.</p>

<p>Both halves were shot with the 55-300 lens, on tripod with VR off, wide open aperture, focal length 55 mm, ISO 100, with back button focus on, set to C and dynamic area, using the IR remote. As noted in my original post, despite back button setting, the Live View side refocused. Aside from switching to LV, nothing else on the camera was touched. </p>

<p>e.t.a. posted before Chuck Fan's post above. Everything he says is what I have always assumed, and since speed is no factor in a tripod shot, it has always been my assumption that, if there is any error, it will show in the viewfinder view and the Live view will be better. And yet, I keep trying shots and I keep getting these anomalous results. All I can conclude is that the decision of what is "focused enough" is more finicky in VF than LF modes.</p><div></div>

[5711]

<p>pick up a camera like a d4s or a 1dx and test what you are talking about. <br>

lv af lacks behind by 10000 lightyears,...at least.<br>

you posted alot of misleading things that are not true.<br>

learn to understand autofocus.<br>

please...do yourself a favour and read posts regarding autofocus on the link i posted.<br>

for once</p>

<p>od damn</p>

[rodeo_joe1]

<p>Norbert, viewfinder (or rather camera-body) AF is affected by lens AF errors - hence the provision of AF fine-tuning. LiveView focussing isn't affect by lens errors, and I agree that LiveView focus <em>ought</em> to be sharper, or at least more reliable than VF focus. Although with my D7200 and 18-140 VR Kit lens there's not much in it.</p>

<p>Matthew, does your 55-300mm lens have VR? I've noticed that VR seems to be less effective when Live View is enabled, and to be honest I don't rate Nikon's VR very highly at all. It's certainly not a patch on Tamron's VC stabilisation IME. The viewfinder image seems to jump visibly just prior to shutter release with my D7200 and kit lens. There's also an unacceptable delay between shutter press and shutter operation when LV is used and magnified. Magnified LV also fails to autofocus properly in some cases. I'm hoping a firmware update might be in the pipeline to fix this.</p>

[Matthew Currie]

<p>Norbert, I have read a great deal about auto focus. I understand the difference between the two kinds of auto focus.</p>

<p>I'm not sure what I posted that is not true or is misleading. Perhaps you could enlighten me.</p>

<p>I have made a number of pictures, using a couple of different lenses, on a tripod, in which the only variable is whether I am using phase detect or contrast detect auto focus, and whatever processing the camera does with the result. When there is any difference noted, the result from phase detect focusing is usually better focused than that from contrast detect. Phase detect autofocusing, at least on this camera, has been consistent and accurate with the AF lenses I have. I am guessing that the camera's decision of when focus is "good enough to shoot" is what differs here, and that it is more demanding when it's set to phase detection.</p>

<p>I don't have any problem with this, but I thought it interesting, because whatever Nikon is doing to make this relatively cheap camera perform well would appear to be effective.<br>

e.t.a. for Rodeo Joe. I tried the 55-300 with and without VR, and there was no discernible difference. I used Live View without magnification, and used the AF provided with each mode, back button for some, normal and self-timer or remote for others. As mentioned, back button focusing does not function with Live View and remote on the D3200, and it always refocuses when using the remote and live view. <br>

I have not noticed a jump in the 55-300 when using VR on the tripod - it's supposed to have tripod detection, and it seems to work all right. The 18-55 kit lens has a very distinct jump. </p>

[sebastianmoran]

<p>Matthew, you are correct, live view should be better. <a href="http://www.lensrentals.com/blog/2012/07/autofocus-reality-part-1-center-point-single-shot-accuracy">See this article</a> for a very careful test by people who know how (Roger Cicala at LensRentals).</p>

<p>In your sample, it's not. That's very confusing. There is something going on here we don't understand. One thought: get the lens way out of focus before each shot, so the AF in either case has to move the lens a lot. </p>

<p>Do you have a prime lens? Or, with your zooms, I suggest trying the same experiment somewhere in the middle of the zoom range, not at the end. Especially not at the long end where the lens isn't as good.</p>

[Matthew Currie]

<p>Sebastian, I'm sorry to say I have no prime lenses in AFS, so cannot test with one. I have tried the two AFS zooms I have at various focal lengths, and the 55-300, at least, exhibits much the same thing at both ends and the middle. The 18-55 has enough depth of field at most settings to make it hard to see. It seems to focus a bit better in phase detect at 55 close up. By the time you get down around 35mm. it's a wash.<br>

I did get the lens well out of focus, and the behavior was the same. Actually, even when the lens is in focus after a viewfinder shot, the LV focus goes into action and defocuses quite a bit in both directions before settling back. It's odd, but I suspect what is happening here is that contrast detect focus, which oscillates back and forth and zeroes in on the focal point, stops when it comes within a certain range of "good enough," while the phase detect focus, which heads in one direction and does not hunt, comes closer before "good enough" is reached. </p>

<p>Interestingly, I have read several of the Lens Rentals articles on AF, and this is one of the only places where I've seen any hint that my issue (I can't call it a problem, really) is not unique, reporting it in some newer Canons. </p>

[pete_s.]

<p>When the camera is using liveview to focus it should be more sensitive to error from focusing in low light. Low light or low contrast means it's harder for the image sensor to determine focus because low light means more noise. And with noise it's hard for the camera to determine if it's in focus or not.</p>

<p>The AF sensors are image sensors too you know but much larger. Some cameras don't have readout of every line of pixels in liveview (due to speed limitations) so it's also possible that the camera is using less than full resolution to find focus.</p>

<p>Either way it's the software in the camera that decides when focus is "good enough". So it is possible that different models are programmed slightly different, just as is the case when it comes to in-camera metering.</p>

<p>So I'm not surprised even if it is not something I have seen or looked for.</p>

 

[Ian Rance]

<p>Live view works best when you zoom in and manually focus on the subject I find. I have tried the live view AF but was not impressed and now use it only in very good light. Contrast detect AF has a way to go yet.</p>

[kari_oinonen]

<blockquote>

<p>Both halves were shot with the 55-300 lens, on tripod with VR off, wide open aperture, focal length 55 mm, ISO 100, with back button focus on, set to C and dynamic area, using the IR remote.</p>

</blockquote>

<p>In this case my question would be, what was the dynamic area actually and where the camera was actually focused. To test it is better to use a single defined af element. </p>

[Matthew Currie]

<p>In the case of dynamic area focusing in this camera, it uses a single starting point, the same as single servo, and tracks from there, but stops when the back button is released. The remote does not reactivate it. </p>

<p>The Live View focus uses an entirely different setting, and for the test I used "normal" area single servo focus without tracking. On this camera the remote always activates it and overrides the back button. </p>

[kari_oinonen]

<p>So your AF areas in the LV and VF -mode were about the same. If not, that could explain something. The target was flat and square to the lens axis - to minimize the effect of using a bit different af areas/spots?</p>

<p>Your example shows exposure difference. Exposure differences may make it hard to tell something about differences in sharpness. Maybe to run the test with the camera in manual exposure mode?<br>

And stay away from any flickering light sources like fluorescent tubes, as they provide natural variability to the light and so to the exposure. Might effect the af too ;-)</p>

[Matthew Currie]

<p>Indeed, there is a little exposure difference, because (I'm guessing) metering is also a little different between the two modes, but is small and correcting in View NX2 does not change the focus. </p>

<p>It's likely true that the difference between modes could be minimized by making the conditions ideal, with better light and so forth, but what I find interesting, and what seems worth noting here, is that the improvement would be to make contrast detect focusing work as consistently well as phase detect. That suits me fine, but I found it a little surprising. </p>