SCoombs
Offline
• •
Upload & Sell: Off
|
 p.8 #10 · Z8 extremely unreliable/inconsistent AF in mildly low light | |
ilkka_nissila wrote:
I am sorry for the snarky appearance of my previous post. The reason for this is simply that the OP has already figured out settings that do give good AF results in his scenario (single point AF) but this somehow is not satisfactory even though it apparently works. I might suggest an improvement to that approach by using 9-point dynamic area (it gives a bit of wiggle room so the selected point doesn't need to be held so precisely on target (in this case the baby's eye) throughout the focusing and photography processes. It's a bit less specific but because the baby skin is smooth it's unlikely to be much less accurate (the skin won't give much of a contrast to focus based on).
The reason subject detection doesn't work well in this scenario is that the AF system requires contrasty detail and the baby's face has none apart from the eyes which are small. When a phase-detection sensor hits right on the eye and is held continuously exactly on the eye for the duration of the focusing, it'll work. When the eye happens between phase-detection points then there is no signal to focus based on and even though the subject-detection process has identified the subject correctly, the camera will focus on the nearest area with contrasty detail which is the pizza. Now, sometimes it'll focus on the eye and sometimes it won't. The pizza tends to win because it's big (so there is no way there is not even a single phase-detection point over the pizza) and it has multi-scale detail (both coarse and fine details) so it is a reliable target for focusing. The camera's autofocus system primarily needs contrasty detail to focus on and something that is reliable over the course of the focusing operation and doesn't "turn off" because of misplacement of the focusing points, otherwise it cannot work. The subject-detection process can only give suggestions to the focusing system but the need for suitable detail to focus based on is necessary. This is nice because if there is no face it'll still focus on some detail and if there is a face it'll focus on it (or the eye, more specifically) when there are phase-detection focus points directly over those features.
If the baby were replaced by a bearded man, the camera would likely focus on the face nearly 100% of the time in subject-detection mode. This is because the bearded face satisfies both requirements for focusing: there are vertical lines, and those lines constitute rather a big area (in a head and shoulders shot) so the signal is extremely reliable, and finally the beard is within the area which the camera's subject-detection process can identify as a face. It won't, however, consistently focus on the iris, because the iris fails to give a large and consistent (over time) signal to focus based on.
When I'm talking about a signal I am really talking about the two 1D images that the phase-detection sensor gives to focus based on. If you switch an f/4 zoom to an f/1.8 prime, the contrast (amplitude) in this pair of 1D images increases because the f/1.8 prime has higher MTF, and the noise decreases if you shoot both lenses wide open because the f/1.8 lens projects gives more light on the sensor, so the noise is approximately halved. Increased contrast and decreased noise both contribute to the eye giving 1D images with higher contrast-to-noise-ratios and so this is why the 35/1.8 can focus more reliably
on the eye than the f/4 zoom. The pizza is a reliable target for even a worse lens because it has high contrast in a large area of the frame and at different spatial frequencies (so even an out-of-focus image gives a reliable pair of 1D images to align by focusing). The reason Canon works more reliably here is that there are no gaps in their phase-detection sensors; all photosites (apart from the very edges) are available for this purpose, so there are always some on the eye. With Nikon, there are gaps between the PDAF sites and if the eye is small enough to fall within a gap and if this happens with both eyes (by chance) then the camera will find the pizza to focus on even though it has correctly identified the face as the subject. It still needs the reliable detail to focus on, otherwise it cannot successfully autofocus. Now, one can debate Nikon's choices
in regard to how the algorithm makes these decisions but there is a logic to it even if it is not the logic that the OP would like it to have. In some situations Nikon's logic works to the advantage of the photographer (when photographing a fast-moving athlete it is better to focus on the texture in the outfit rather than the face that doesn't give a clear signal to focus based on due to the eye moving on and off specific PDAF points and sometimes being between those points, even if the focus won't be perfectly on the eye it'll still be on the athlete from which it is easy to make a correction when the focus is on the eye).
Focusing on baby faces is difficult and I would customise a lens button to override the normal subject-detection-enabled focus area mode that you're using with 9-point dynamic area. When the face-detection (usually I use C1 or C2) fails and focuses on a bright and contrasty vertical line that is not a face, I'll override it with 9-point dynamic by pressing the L-Fn2 lens button if it is available on the lens (e.g. 70-200/2.8 and 24-70/2.8 Z S lenses have it). Other lenses have just one L-Fn button or none, alas. Anyway this is how I am able to avoid or work around these problems that the subject-detection modes have and get good
results with minimal fuss.
I'm not a camera designer and so I have no inside information on the camera's inner functioning but as I've said I've shot more than 1 million frames with Nikons and photograph many different action subjects and I often check the focus points when editing and selecting my images on the computer. On the D6 the camera will actually show two different focus points one is the focus area that is selected and the other is a white box showing the subject identified by the camera's subject detection process (unfortunately I've not figured out how to display the white box outside of the camera's replay feature). These two points often do not match the subject focus (sometimes they do) and when I've looked at numerous images I've come to the conclusion that the subject-identification process runs in parallel with the focusing process and the subject-identification process can only suggest an area it has identified as a subjet for the camera to focus on but the focusing itself takes time, and often by the time the focusing has been achieved the subject is identified elsewhere (in fast action such as figure skating). However, the subject-identification process does improve results over plain group-area (in the D6) or wide-area or auto-area (or 3D tracking) it just has its own limitations and so does the camera's focusing process. You just have to accept the limitations of these systems and work around those to get the best results you can. The systems are completely amazing compared to what we had some 20-30 years ago, even if they're not perfect and you still have to adjust settings from time to time.
What bothers me is that the OP posts a highly confrontational and in my opinion misleading title to his thread, and then even though there are solutions to the problem, he does not accept them as solutions. I cannot help but think if the goal is just to create controversy rather than make photographs. That's why these debates get heated. Neither is the camera perfect nor is it unworkable. You just have to find the solutions to your particular shooting situations and use those solutions. Different people have such diverse shooting needs, there is never going to be a camera that will do exactly what you want it to do in all respects, unless you design and manufacture it yourself. ...Show more →
Look, I am NOT trying to be confrontational; I really am not. Now that you raise the tissue, I do see that the title could have been presented in a slightly different way. I have been struggling with extremely unreliable and inconsistent AF in low light and am looking for help. I titled the thread accordingly. Truthfully my memory was that I thought I had tried to be diplomatic in the title, which is how I normally approach things. I would say this title is a bit more direct than it could have been.
As far as the rest of your post goes, I have to say I'm somewhat heartened to see you apparently agree that the subject detect isn't working well in low light. You also give a bunch of reasons why you suspect this may be. Whether you're right or wrong in all of this, the fact is that so much of this thread consists of people insisting I'm wrong and that the subject detect DOES work flawlessly even in the dark. I had thought that in your previous post you also essentially said as much, but perhaps I misunderstood. Either way, as I see it you're sort of in the minority (here of FM anyways) in agreeing that subject detect probably didn't focus accurately in these conditions.
More importantly, I noonger think I have a solution. Perhaps you did not see this in the mix of all the other posts, but Friday at an event I found even single point and dynamic area modes to be inconsistent and unreliable. I did say in my OP that in my testing at home these seemed to work at a higher clip, but Friday night I found myself having to check every photo and take every shot 3 orn4 times to be sure I was getting them with single point and dynamic area. I shot somewhere between 1000 and 2000 photos and will be providing around 150 or 200. Many of the rest are out-of-focus or else duplicates I had to take to ensure I got anything. Of the ones I do have, I'm sending more through Topaz (something I hate doing) than I ever have before because truthfully only a tiny handful are actually truly sharp and in focus as they should be.
|
When Nikon has released a camera that fell short of expectations, they’ve gotten pilloried for it. See: the Z6 with only one card slot; the Z9 & Z8 birding AF in early firmware. Instead you have documented case after documented case of people saying they’re happy with the cameras, and up to the task of shooting in low light. You’ve got Richie and Approaching The Scene demonstrating AF performance in real time with Live View recordings - not in low light, but their transparency about the performance is thorough. You’ve had well reported and documented instances of the AF on the Z9/Z8 being disappointing; and then the follow ups when firmware has led to improvements. I’m sure plenty of people have figured out ways to avoid taking photos successfully, and quickly vent their frustrations on the internet, but this has no bearing on the effectiveness of a camera in competent hands.