Beware, this is an extremely long comment. I figure you know some or most of this, but the audience watching this video might not. Contrast detect AF, especially in continuous AF, needs to do a sort of constant "breathing" because it detects edges, and it can only detect the difference in those edges during a defocus. If it doesn't re-check for edges, it has no idea whether or not it's in or out of focus, and the only way to determine that contrast difference is to force a contrast check, which makes a slight wobble of the lens as it quickly makes a minor focus adjustment. It's usually looking for horizontal detail, and that's often troublesome when you have very small points of light. Contrast AF is not aware of depth, or the amount of defocusing, and unless it does a re-check (the quick hunt) it doesn't know whether or not it's in focus. There are a few types of AF: 1. Range-check. This is done using an external measuring system. The first AF cameras used sonar reflections to determine the distance to the subject. This did work, unless you had something blocking the reflection or very long distances. This was introduced by Polaroid, but was quickly eclipsed by other systems. Because it's not through-the-lens, it's only useful for fixed-lens cameras with no filters attached and it's limited to nearby subjects. You could use another measuring system, like lasers, but the principle is still the same. 2. Phase detection. This is what was introduced in the early 80s as a through-the-lens method. While Pentax had introduced a contrast-type AF camera with the ME F in 1981, the first phase detect SLR camera was the Minolta Maxxum 7000 (or Dynax 7000 in Europe, or Alpha 7000 in Japan). Featuring its through-the-lens phase detect AF module and built-in body motor (instead of every lens requiring a motor), it was the first system camera to popularize AF. The trademark was "Only the human eye focuses faster," and, well, it was true. Its single AF point is laughably quaint by today's standards, but Minolta did put together the first really serious AF camera system. Unfortunately, they infringed on Honeywell's phase detection AF patents to do this, and the resulting court case essentially crippled the company and set it on a path to merging with Konica. All of their misadventures with the xI series also sealed the deal. But that's getting ahead of ourselves... Phase detection works by comparing the phases of two out-of-focus areas and determining the difference of focus (backfocus or front focus) and exactly by how much. It's not that much different in practice to using the split-prism in a manual focus camera's viewfinder. The camera then moves the lens focus group by exactly that amount. It is swift and sure, but it can also be fooled! All SLRs since the Maxxum 7000 use this method, just in greater complexity. Phase detection is great for continuous tracking AF performance, especially when combined with predictive AF algorithms. By comparing the in-and-out of focus areas not by contrast but by phase difference, the camera doesn't have to second-guess itself. It just makes the move based on the detected focus difference. The downsides of phase detection are: 1. The modules are not in line with the film plane, meaning you can get distance measuring errors if the module is not in perfect collimation. 2. You're limited to where incident rays fall on the exit pupil of the lens. Because these systems use semi-silvered mirrors to steal light, you often get coverage that sticks to the central part of the frame, at best. The farther you get away from the center axis, the worse the phase comparison performance gets. 3. The phase sensors are limited to detection at certain F-stops and contrast directions. Some cameras have sensors with multiple levels of sensitivity (e.g. most Minolta cameras post the Maxxum 7, which have a double-cross f/2.8 sensitive sensor, along with f/6.3 sensitive sensors), but you only get that sensitivity boost on that one sensor. You can't AF with anything at a narrower aperture, and because the lens has to focus wide-open, you can get focus shifts when stopping down. 4. These systems can be fooled by very near subjects. The human does need to give it some help (e.g. using the joystick to position the AF point or using a focus limiter to ignore certain ranges). Next, is 3. Contrast Detection. This is what's used in video cameras with CCDs, and most point-n-shoot digicams of the 90s and aughts. Contrast detect works by finding edge contrast. Go into Photoshop and use the "Find edges" filter and you'll see what I mean. The advantage to this is usually that it measures right off the sensor plane, so you'll never get front/backfocus errors, and you can use it at shooting aperture, which means you can dodge focus shift (most of the time). By making a quick snap of the focus, the camera can find edges in the image, and then once the edges reach a certain acutance, it locks focus. This is great for single AF (lock on and then don't refocus). However, this is bad for continuous AF, because it doesn't actually know anything about the direction or amount of defocus. It only knows about edge contrast. Some manufacturers (like Panasonic) have highly tuned this with their Depth from Defocus contrast system to improve continuous AF results, but it's still a bit weird and behaves differently from the phase AF-C that most people are used to. You CAN combine these AF systems, especially in mirrorless camera bodies. Sony has lead the field in on-sensor phase detection systems, where certain pixels are masked to behave like phase-detect sensels. By having the phase detect on the imaging plane, you ditch the front/backfocus problems, and can have a much wider AF point spread too because the semi-silvered mirror is gone. By combining the on-sensor PDAF with contrast detection, you can get a best-of-both-worlds hybrid system where phase detect can do raw defocus/distance measuring and contrast detect can get edge acutance. This is what's setting the world ablaze with Sony's mirrorless cameras, and because other manufacturers are using Sony sensors (like Fuji and Olympus) they too can enable similar features in their cameras, albeit with their own implementations. On-sensor PDAF can also be mixed with off-sensor PDAF, as seen in Sony's a99ii, but it's a rarity amongst cameras for a variety of reasons that are beyond the scope of this comment. ALL AF systems are limited by light and contrast range (just different kinds of contrast). Very low light areas or areas with no detail (e.g. flat surfaces) will lose focus. This can be worked around (AF assist lamps), however the behavior of hunting changes based on focus methodology. Phase detect hunting tends to go towards extremes (fully in or out) while contrast wobbles and steps slowly. A wide AF array should be able to see large groups of objects, but without edges (contrast) or lines (phase) there's just no detail to check. Some cameras' wide AF methodology can only check for areas inisde one point (determine a point) while others will use multiple points in the plane of focus to keep the lock. Continuous AF is best for a moving subject, this is a case where using AF-S, back-button focus, or a program/camera that can lock or hold focus is better than just sticking with continuous AF. There's also Canon's Dual Pixel AF which is a different physical of on-sensor phase detect, but works towards the same goal. Aptina (another sensor manufacturer) also did its own on-sensor phase detect system used in the Nikon 1, but they have since cross-licensed their patents to Sony and pulled out of most of the market. There's a basic primer on AF systems, how they work, their gotchas, and their strengths. I forewarn you, going down this path is a dangerous one. You will start attracting camera nerds who put brand above all else, and they are not for the faint of heart. However, your videos are great and if anyone can do a straightforward explanation of how all kinds of AF works, I believe you are more than up to the task.
"...audience watching the video might not." Hey, I'm here *specifically for* this level of detailed tech info and geekery! To paraphrase Tech Connections Guy, "spoiler: huge nerd." Thanks for the edification. 😁
Please, if you want do some research and make a very long video explaining the autofocus on old camcorders, modern cameras and mobiles phones, it would be amazing.
Pretty soon I'll be starting some videos on camera tech, so I'll be sure to include a video on autofocus! I'm sure there's a lot of interesting stuff to uncover
Before someone brings it up; yes there are more than just 2 airports in Chicago, but O'Hare and Midway are the major ones that you or I would be travelling through.
I remember old Meigs Field for it being the default airport in Flight Simulator. Years later, I read about what happened, how it was demolished almost as a covert op, in the middle of the night and without notice. Sad.
I've always felt that MDW was a bit of a harsh landing... I like to sit near the wing and watch them "panic" to stop the plane while the engines reverse. I'm guessing the runways are a bit small. Still, living near it, it's the only one I fly out of.
This is part of why almost all "real" video cameras have a setting (often a dedicated button) to lock it on infinity focus. In fact, specifically in regard to a nighttime scene with "bright points of light", I've seen some camcorders that have a "Fireworks" preset which locks it on infinity focus and reduces the exposure to prevent the fireworks from getting blown out.
It's very frustrating to me how smartphone cameras are often so close to perfect, but with some serious annoyances or flaws. What I really appreciate about the camera on the OP3 is that its image stabilization is really good, even in this low light. I didn't do anything at all with this footage, and you can see during the landing that the phone was moving around pretty violently (pay attention to the window near the edges) but it coped very well. Smartphones have the benefit of so much software but sometimes implement it in strange ways.
Technology Connections 2 On my OnePlus 5T you need to tap and hold on the point you want to focus and eventually you'll get a message telling you the focus has been locked. I'm not sure if it's the same with the OP3 but I thought it would be worth a try
I really wish I could just do manual focus on my S5. I'm running Android 6 and I've tried advanced camera apps but none of them work. It's especially annoying when trying to focus on something up close. I know the hardware is capable because it will be in focus for a split second before autofocus decides it'd rather make it blurry.
I guess I just have to think about some random thing what interest me but never looked really into it, and wait for a while because this and the main channel delivered the answers every_single_time for the past few months. Stop following my thoughts, man!!! Actually, please continue.
They’re not “points” of light. They’re blobs. I suspect, as the light spreads, it leaves few actual sharp edges. It’s more like the dark and light areas are fading into each other rather than creating contrasting features that the AF system can use.
That seems like a software issue to me. When you tap the screen to focus there, it looks like the camera continues scanning that area to maintain focus. Once that area has no more contrast, the camera starts hunting. Cell phone cameras have to be dead-simple, so that people who -think vertical video is good- are novices at photography can use them. That behavior might be a compromise to get better ease of use for novices. Also, I don't know if this is true for contrast-detection AF, but with phase-detection AF, you need detail along a line for the AF to work. Focusing an SLR on a point is difficult unless there's a ton of light.
My understanding of this problem is that when a pin point of light comes out of focus it becomes I high contrast circle with reasonable crisp edges that are easily detected by the camera making it think it is in focus. I'm sure once AI get better in camera, they will take into account single points of light, but for now they do not. Take a screen shot of a still frame when the camera is out of focus on a dot and you will see it is a nice circle, or at least in the shape of the aperture of the camera and has a high contrast edge.
Agree with the assessment. Another thing is dynamic range. Contrast AF computes the sum of squared differences between neighbouring pixels (or a variation). In a dark scene, point lights are over-saturated. Suppose saturation level is S, a point light has actual level of 9S, it would still register as S. If it defocuses to a disk of 9 pixels, each pixel would still be S. The circumference increases by 3 times, and the sum of squared differences increases by roughly 3 times. Thus the AF often prefers somewhat out of focus blobs.
I had the same thought. The bokeh of a camera doesn't produce the same result as a blur, it actually makes a distinct circle on point lights. Bigger circles could actually have more high-contrast pixels than when it's in focus.
If you tap and hold in the camera app you can lock the focus and exposure, I tend to do this often when filming with my phone's camera since it bounces everywhere otherwise.
I'm going to guess that the primary reason Southwest, as a low-cost airline, uses MDW was because as they were probably able to negotiate lower landing fees compared to O'Hare and the fact that since their fleet mainly consists of 737s means that the shorter runways are adequate for their needs. In the UK, London Stansted is a similar case - it's a hub for Ryanair (an infamous low-cost airline) while most of the other major airlines use Healthrow and/or Gatwick instead
The only reason I can think of for the focus-hunting is that the MASSIVE difference in contrast confuses it. "Wait, that can't be right, I should probably refocus to make sure." Then the few non-points-of-light in the frame also confuse it. "Huh, those points of light are clear, but these others aren't, maybe my user wants these in focus? Oh, they won't focus, because they're not actually points of light."
Try the OpenCamera app -- it has focus lock even on videos but it has no other fancy functions such as filters, video cropping, etc. Used that and never had any problem with it losing focus.
Penurious Sierra yep pros use manual focus. If you want to make pro videos dont se a mobile phone. Ps my camcorder is always set auto everything cos I'm not a pro. But when I do dabble you can get great results with manual focus.
Penurious Sierra That is one of the reasons I have it on my phone, the focus and exposure lock options. It can even rotate and crop automatically to keep the horizon level. And timelapse stuff
Another consideration might be that autofocus once "locked on" will only "fully" reevaluate the entire image periodically, but relies on pan/tilt data from the video compression chip more often to track the movement of precomputed contrast boundaries as a lazy refocus mechanism.. it could be designed to do it that way more often to reduce computational cycles... with so little variation in these scenes pan/tilt errors could add up enough it could easily get confused, lose its contrast boundaries by a pixel or two, and have to start from scratch again... if I had to design it, I’d consider that as one autofocus mechanism.
My old Sony handicam years ago had a switch for auto/manual focus. The nice part was that it also had a momentary toggle - if you pulled the switch all the way down it would focus to infinity immediately and then it would be in manual mode. If the auto focus was being screwy you could just grab this easily and tap it to inf and switch back to auto - it would usually fix the problem right away.
I know for sure that the aircraft in the video was a Boeing 737 of some sort with the CFM International CFM56 engines. I can tell from the sound of the engines when it touches down on the *RUNWAY* . It's not the "tarmac", at least not called that, even if it's made from that material.
My understanding is that these sorts of autofocus systems require contrast on a particular axis. So they can focus on something like vertical blinds, but not horizontal lines.
I’ve figured it’s because with points of light and lens blur points become circles with a high contrast edge. So being out of focus optimizes the greatest area/number-of-pixels with high contrast.
PS I figure phase detection focus wouldn’t have this issue; it would have other issues, maybe lower likelihood of the focus pixel having light go to it.
My guess would be that the camera has difficulty with depth perception, which is what focus really is, focusing on how deep(away) something is. So, with only points of light and not a continuous source of depth information, it would make sense that the camera would struggle to focus.
I've noticed that my Sony HDR-CX760 has no problem with scenes like this, but I think there is something else going on (though I'm very likely wrong). That's why I use Canon video cameras sometimes because external focus sensors (iAF) that everybody seems to mistake for an extra light or a stabilizer sensor.
I would imagine that the problem it's having is that the points of light that it's trying to resolve are all at different distances, and therefore it's trying to focus on a new point of light every times it focuses on another. I don't know crap about how cameras work, but it sort of makes sense to me.
Two reasons i can think of. 1 Because the sensor is dimly lit, there is quite a lot of noise that the af algorithm will struggle with. 2. A point light source with most lenses will when seriously out of focus show nicely round bokey balls which might throw off the algorithm
I think it does not detect contrast between pixels but tries to find straight lines in picture. The simplest (and very fast) algorithm is to use convolution function and then see at which focus setting there is greatest amount of them visible.
This is a sheer guess on my part. But I believe it may have something to do with the fact that the points of light are moving in the frame. In other words if they were static it would be in the number same pixels all the time. Yet since they are moving they could be in X number of pixels in one frame and Y number of pixels in the next leaving the system to believe that is suddenly out of focus.
It was my understanding that it only looks at a small region, first of all. Secondly, at least on a camera system I worked on years ago, that you would focus by finding a sharpness value of a frame, blur the image in the system, and get the sharpness value for that. compare that. If it is similar, it’s likely already out of focus. Then adjust the focus of the lens and repeat while adjusting that. The highest sharpness value is the proper focus, probably. Then every so often, you need to run it again to make sure you’ve maintained focus. That’s how our system worked anyway.
Some cameras use on-sensor phase detection, which isn’t so great in low light. I know canon’s dual-pixel cmos autofocus has issues with very low light focus because it’s phase detection based. The oneplus 3 uses a 16mp sony PDAF sensor, which stands for Phase Detection Auto Focus. Looks like that’s why you’re focus hunting.
Ooh, really? I never knew phase detection was in use in smartphones. Research fail! BUT, given the tremendous ability of a smartphone to use software, it surprises me that there's no analysis taking place on the software side to minimize the focus-hunting. Perhaps we'll move towards that some day--best of both worlds!
Some do. This is what Sony's Hybrid AF does in their mirrorless cameras which use on-sensor phase detection, and I presume this is also available on their small sensors used in cell phones. However, it is a function of image processing, so if the AF methodology of the phone's image processor doesn't do it, then they would have to implement it. You can force those cameras to do different focus methods based on focus modes, FWIW.
My guess is that autofocus algorithms contain an escape algorithm in case a small insect were to fly right in front of the camera. The algorithm would start tracking the tiny insect, and should the algorithm try to always obtain maximum contrast, there is a local maximum of best contrast around that insect and the AF algorithm would not be able to focus on anything behind. The same phenomenon would happen if trying to film anything through a window (any tiny speck of dirt on the window would steal the autofocus and never release it). The algorithm probably detects the size of objects and ignores them if the object is too small. Making the camera's autofocus algorithm ignore isolated point lights.
Maybe it operates after a low-pass filter to avoid false triggering from sensor noise? In that case it could be possible that the points of light just don't cover a wide enough area to be significant enough for focus lockup
I noticed this too, and I think I figured it out... Maybe. Some cameras use both a laser/IR rangefinder, and as a fallback it uses the contrast detection, and they can conflict a bit. What I have found is that most phones (#ButNotAllPhones #ButSometimes) that work with alternative camera apps that have a 'lock focus to infinity' setting, which works for these types of situations. Another option is 'fireworks' mode, which should do roughly the same thing but also bring some increased brightness... I use Open Camera on Android and it works pretty good. YMMV. But the real reason is that it looks for areas of the MOST contrast, so the tiny little points of light are probably 'too small' or considered noise to be used as valid things to be focusing on.
I would say it's because the iso is probably as high as your camera can muster there and that make for a noisy signal. So the program might read the nois around the light as right next to the lense and that would make it pull back to it. If that's not it I blame the window.
Maybe there was something on the window? Sometimes there's a little dirt, speck, etc, and the camera tries to focus on that. Or a reflection from inside the plane. It could also be what others have already suggested: the camera has certain point[s] that it uses for autofocus, and when there's nothing in that particular point, it struggles. Though it would seem trivial for a modern smarthphone to scan the entire image with software, especially since everything was at infinity.
I believe this happens because the camera defaults to trying to focus on a small square in the center of the screen. When you tap, you're suggesting the best place to compare contrast. And when the image changes beyond a certain threshold, it refocuses either back to the center or at the place you tapped depending on the camera app, and by then the light source has moved and it's just darkness its trying to focus on. No contrast, as you know, means poor autofocus. I don't think the software considers the bulk of the picture when focusing, just a tiny sample.
Could it be related to how out-of-focus lights look like in cameras? Differently from a gaussian blur, lights look like little disks with relatively sharp edges on camera. These edges may fool the AF algorithm, leading them to prefer the out-of-focus image to the sharpe one, since the latter has much smaller edges (although slightly sharper). The fact that bright light are probably clipping, and therefore the brightness level in the sharp image is not much higher than in the out-of-focus one, may help too.
Thank you for posing the problem. I have been going crazy with my Oppo A3S mobile jumping in and out of focus on videos and thanks to the learned kefkafloyd, I understand why. I have downloaded an app mentioned by one of the other replies and I am looking forward to seeing if it works. Love your channels. Cheers.
Maybe because contrast AF works with the unfilteted, damned noisy raw image data with a lot of "fake contrast", not the heavily noise reduced processed data you seebon screen/video?
Thanks for calling ATL by its full name ;) Interesting note, unlike Chicago O'Hare, which is gerrymandered so as to be completely within the city of Chicago, Hartsfield-Jackson Atlanta International Airport has some concourses (T, A, and part of B) in College Park, GA and others in unincorporated Clayton county...oh, and a small sliver of the northern end of the airport where Delta is in Atlanta proper. This led to a tax dispute on whether the city of Atlanta (which owns and operates Hartsfield-Jackson) owed College Park back taxes over liquor sales (they do) and whether Atlanta can collect taxes on booze going forward (they can't).
Do phones actually use contrast detection autofocus these days? I keep hearing about Focus Pixels and laser autofocus and different exotic techniques that supposedly replace it.
This video ruined my relationship. I was watching this video with my girlfriend, and when we saw the extra S, we both laughed really hard about it. The thing was, I couldn’t stop mentioning it. I was making references to it all night, hours after we had finished the video. Get this, I woke her up in the middle of the night to say “S.” She grabbed me by the hair and then made me sleep on the floor after I did that.
Personally I just came back from Houston recently and took some very similar shots coming back at night. I have always used Center point focus and have never had an issue. Something to be said about using grid focusing in certain scenarios but this is probably not one of them. When I take a picture I tap my focus point so the phone knows to focus on what matters to me.
As someone who grew up in Schaumburg and always flew in and out of the Chicago area, can I just say fuck Midway? No, seriously, the one and only time I flew out of there on Southwest Airlines my bag went to fucking Hawaii when I ended up going to Phoenix. I didn't have a suit to wear for my great grandfather's funeral because of that shit.
I've always understood that contrast auto focus isn't as accurate or as fast as phase detection auto focus on SLR cameras, also the auto focus region on many cameras only covers a small area in the centre of the sensor, so if the centre of the image has large dark patches auto focus may have trouble.
Does the same thing happen with other scenes that include a lot of solid color (besides darkness)? I suspect the root cause of the "hunting" is that darkness is the easiest way to flood the sensor with a single "color". Outside of scenes of darkness, fields of truly solid colors are probably fairly uncommon. This is probably an example of "optimization for the common case".
Here's what I think is happening: your phone (OnePlus 3 isn't it?) uses Contrast Detection AF, like you said, and it needs contrast, but it needs contrast *detail* . The low light scene means that it can't see surrounding detail at all, so it defocuses a lot instead of in small steps to measure the brightest region's smallest size and stick to that focus. The thing is that the AF depends on the metering to tell it where the brightest spot is,and since that is constantly changing, the system constantly has to relocate the brightest region, and measure multiple focus stops till the smallest size is achieved. Then by the time this process is over, if the metering system tells it that there is a new brighter spot in the frame, or that the brightest spot had changed intensity, the process has to be started all over again. The main sticking point is that there are spots of bright light in frame, but all around those, is simply featureless darkness so the region could already be in focus and this means that the entire focus range needs to be checked to compare the spot sizes.
I agree completely It is baffling. My latest fireworks video is better than some I have done, but it still isn't very good. What is even more baffling is I thought you were the guy to tell me how to get it right :P
I think it boils down to the camera simply not looking for focus where the details are. And sometimes it can mistake the circle of confusion for an in focus shot mor than the actual in focus shots. Also. Just in general. If the distance is not expected to change significantly... It's generally preferable to simply get focus and then just turn off the autofocus. That way there is no focus hunting at all.
Imagine a lattice of small lights in an unlit background (like nights). Imagine that you can increase or decrease the areal density of those lights. Lets say we hv a way of measuring contrast (actually a metric to indicate contrast). Now imagine two extremes- 1) with only one light. 2) very densely packed lights which are still resolvable. First case would hv a very low contrast metric and second case would hv a high contrast metric even though for humans, it seems reasonable to term both cases being equally contrast (actually the first one might even be judged higher). So it follows...
Depending on what autofocus technology your camera employs, it may well be trying to focus on the pane of the window glass itself. Experienced photographers usually shut off autofocus and manually focus to infinity when shooting landscapes thru windows for this very reason.
Jeez... It's been a LONG time since I've seen a jet use the whole runway before taxiing off... Even at COU, which has a 6500' runway, we usually only use about 2/3 of the strip...
I *never* use auto-focus if I can at all help it! I long for the days where focus was achieved using MANUAL focus-ring, NOT 'up' and 'down' buttons to set it!
Well, in the plane there are two problems. Part of the basic answer to your question is lens aberrations. Exacerbating this in many situations, as here, is the fact that there's yet ANOTHER (potato quality) lens between the camera photoreceptors and your subject matter: the window. From a foot or two back, the phone might even successfully bring the window into focus. Well, not the window, but all the imperfections and dust and etc
I wonder if you'd have the same autofocus problem outside the cabin. Like, is it possible that the camera is focusing on the window due to its imperfections? I have that problem all the time with older phone cameras.
I think that's because not all the points are at the same distance. But the contrast of all the points (points are lights) is the same. So the autofocus is trying to focus on all of the points. but that is impossible so it will keep hunting. But that's my guess and I have not looked into it to. (greetings from the Netherland)
android device? you can install a different camera app to allow manual focus in video. Are there any updated properly to support your phone... not sure - any of then allow setting manual focus in video... also not sure
Ive always been frustrated trying to take pictures at night. Only success has been with a tripod. Whenever in an airplane or a helicopter stills and videos are random. Take a ton and toss most.
A simple (and probably partially wrong) explanation would be that the points of light don't take up enough of the image. While a naive average approach would excel at focusing in such circumstances, this is more of a rare exception to how terrible naive average approaches almost universally are for video processing. Whatever algorithm is being used probably intermittently filters the sparse points of light out as background noise or something.
This aviation geek is rooting for more plane videos (and maybe aviation tech!). Also as I watched this video, I wondered what frame rate Premiere showed you? Everytime I use my Moto G5s Plus for video, I seem to get a variable frame rate (non-standard). I've always wondered why cell phones do this. My guess is probably for getting better exposed video. They have the processing power to encode and decode it quickly. Probably much more power than a normal video camera.
As far as I know AF Systems Look for edges, Not points. A single point out of Focus tends to have a very sharp edge around it which confuses the systems.
Thing is, the more in focus the images are, the more chance that the points of light are blown out beyond 100%. 100% white isn't very different from 100% white, whereas if it defocuses and spreads out the light, it can contrast varying shades of grey or yellow.
I have very mixed feelings about it. The Blue Line already gets you to the Loop in 40 minutes for $5 (only $2.50 headed back to O'Hare). It seems a little too much like a benefit of diminishing returns that only the relatively wealthy will enjoy. Then again, if Musk is willing to foot the bill, it seems like it can't cause much harm. Also I did hear that there could potentially be TSA screening at the station in the Loop and you would be let off inside the sterile area at O'Hare. I might even pay extra for that. But my opinions come with the disclaimers that I am a suburbanite with a car that almost never goes into the city to begin with, so I can't put myself in the shoes of those it would benefit most. What I think bothers me most is that it may create a new class divide. I don't think it would be good for people to see the 'L' train as a second-class option, which I can't help but imagine will occur if Musk's train comes to fruition. Here in the States many have a grim view of public transit in general, which I don't want made worse.
It's interesting - I always thought it was only me who was infuriated by this haha... Thinking about it though with 2021 technology - this should be easy to overcome really. The answer? Machine Learning... Apple and Google are doing this all the time, so that when you take a photo of a person it is able to recognise that and adjust the camera settings automatically for the best shot. Sun in the background? No problem, auto HDR... etc. Phones also have sensors in them - accellerometers and gyros to know which direction they're pointing/how fast they are moving, etc. Surely Apple or Google could simply bake in some logic along the lines of "IF the scene is low-light AND there are multiple points of light that look like roads AND I'm moving at over 100mph, AND I'm at an altitude well above what maps is telling me the ground level is at this lat & lon, then assume I'm flying at night and keep near-infinity focus the whole time....." Something like that...
My guess is that the bokeh effect on punctual light sources creates a round shape with pseudo sharp edges (for example: bit.ly/2tdAf4y and bit.ly/2MNrwhE), so, under those circumstances, even when it's absolutely out of focus, the contrast detection believes it's correct. I recall some (many) years ago when some people pointed their VHS cameras to the night sky and zoomed in at stars, and they believed they were taping UFOs, because they saw "circular shapes" flying around.
Oh man, I just realized how much I miss my watch beeping on the hour; it's been a couple years now that I've been a pebble user, and pebble has no piezo.
All that space to build and people choose to cram up together. I would hate to live in a busy place. I'm so lucky to be able to take a short walk and have no people or buildings around me
Tangobaldy Chicago’s reason for existing is that it is where several transportation routes cross in the middle of North America. People are clustering close to those. Yep the emptiness close by is just fine with me too.
Cities exist to bring people and products/services close together. I can have amazon deliver most things the same day, I have a game shop about a ten minute bus ride down the street, right next to my community college. Driving a bit farther I can get to old town where there are several nice Italian restaurants, an apple store, several clothes stores or whatever (clothes = boring), a really good build-your-own ice cream shop, and a dozen other places I’ve never been inside. I too enjoy spending time away from the chaos and in nature, but I value the access being in a city provides as well. (We also have a mountain to hike on if one wants to be away from the city for a while, however it’s California so for most of the year it’s basically just covered in dust and dead plants, and is quite hot.) On the other hand, as delivery gets cheaper and work moves online I can certainly see more people moving away from cities
Mobiles (cell phones in your silly country) also have sensors to know how far away the item is to focus on. As you're looking through windows that will likely not be focused (see what I did there) on clarity but strength, so extra sheets on them you won't notice, this will no doubt be picked up by the phones sensor. The beam will be hitting the glass and coming back as though you're filming something super close and explains the massive jump in focus at times.
Beware, this is an extremely long comment. I figure you know some or most of this, but the audience watching this video might not.
Contrast detect AF, especially in continuous AF, needs to do a sort of constant "breathing" because it detects edges, and it can only detect the difference in those edges during a defocus. If it doesn't re-check for edges, it has no idea whether or not it's in or out of focus, and the only way to determine that contrast difference is to force a contrast check, which makes a slight wobble of the lens as it quickly makes a minor focus adjustment. It's usually looking for horizontal detail, and that's often troublesome when you have very small points of light. Contrast AF is not aware of depth, or the amount of defocusing, and unless it does a re-check (the quick hunt) it doesn't know whether or not it's in focus.
There are a few types of AF:
1. Range-check. This is done using an external measuring system. The first AF cameras used sonar reflections to determine the distance to the subject. This did work, unless you had something blocking the reflection or very long distances. This was introduced by Polaroid, but was quickly eclipsed by other systems. Because it's not through-the-lens, it's only useful for fixed-lens cameras with no filters attached and it's limited to nearby subjects. You could use another measuring system, like lasers, but the principle is still the same.
2. Phase detection. This is what was introduced in the early 80s as a through-the-lens method. While Pentax had introduced a contrast-type AF camera with the ME F in 1981, the first phase detect SLR camera was the Minolta Maxxum 7000 (or Dynax 7000 in Europe, or Alpha 7000 in Japan). Featuring its through-the-lens phase detect AF module and built-in body motor (instead of every lens requiring a motor), it was the first system camera to popularize AF. The trademark was "Only the human eye focuses faster," and, well, it was true. Its single AF point is laughably quaint by today's standards, but Minolta did put together the first really serious AF camera system.
Unfortunately, they infringed on Honeywell's phase detection AF patents to do this, and the resulting court case essentially crippled the company and set it on a path to merging with Konica. All of their misadventures with the xI series also sealed the deal. But that's getting ahead of ourselves...
Phase detection works by comparing the phases of two out-of-focus areas and determining the difference of focus (backfocus or front focus) and exactly by how much. It's not that much different in practice to using the split-prism in a manual focus camera's viewfinder. The camera then moves the lens focus group by exactly that amount. It is swift and sure, but it can also be fooled! All SLRs since the Maxxum 7000 use this method, just in greater complexity. Phase detection is great for continuous tracking AF performance, especially when combined with predictive AF algorithms. By comparing the in-and-out of focus areas not by contrast but by phase difference, the camera doesn't have to second-guess itself. It just makes the move based on the detected focus difference.
The downsides of phase detection are:
1. The modules are not in line with the film plane, meaning you can get distance measuring errors if the module is not in perfect collimation.
2. You're limited to where incident rays fall on the exit pupil of the lens. Because these systems use semi-silvered mirrors to steal light, you often get coverage that sticks to the central part of the frame, at best. The farther you get away from the center axis, the worse the phase comparison performance gets.
3. The phase sensors are limited to detection at certain F-stops and contrast directions. Some cameras have sensors with multiple levels of sensitivity (e.g. most Minolta cameras post the Maxxum 7, which have a double-cross f/2.8 sensitive sensor, along with f/6.3 sensitive sensors), but you only get that sensitivity boost on that one sensor. You can't AF with anything at a narrower aperture, and because the lens has to focus wide-open, you can get focus shifts when stopping down.
4. These systems can be fooled by very near subjects. The human does need to give it some help (e.g. using the joystick to position the AF point or using a focus limiter to ignore certain ranges).
Next, is 3. Contrast Detection. This is what's used in video cameras with CCDs, and most point-n-shoot digicams of the 90s and aughts. Contrast detect works by finding edge contrast. Go into Photoshop and use the "Find edges" filter and you'll see what I mean. The advantage to this is usually that it measures right off the sensor plane, so you'll never get front/backfocus errors, and you can use it at shooting aperture, which means you can dodge focus shift (most of the time). By making a quick snap of the focus, the camera can find edges in the image, and then once the edges reach a certain acutance, it locks focus. This is great for single AF (lock on and then don't refocus). However, this is bad for continuous AF, because it doesn't actually know anything about the direction or amount of defocus. It only knows about edge contrast.
Some manufacturers (like Panasonic) have highly tuned this with their Depth from Defocus contrast system to improve continuous AF results, but it's still a bit weird and behaves differently from the phase AF-C that most people are used to.
You CAN combine these AF systems, especially in mirrorless camera bodies. Sony has lead the field in on-sensor phase detection systems, where certain pixels are masked to behave like phase-detect sensels. By having the phase detect on the imaging plane, you ditch the front/backfocus problems, and can have a much wider AF point spread too because the semi-silvered mirror is gone. By combining the on-sensor PDAF with contrast detection, you can get a best-of-both-worlds hybrid system where phase detect can do raw defocus/distance measuring and contrast detect can get edge acutance. This is what's setting the world ablaze with Sony's mirrorless cameras, and because other manufacturers are using Sony sensors (like Fuji and Olympus) they too can enable similar features in their cameras, albeit with their own implementations.
On-sensor PDAF can also be mixed with off-sensor PDAF, as seen in Sony's a99ii, but it's a rarity amongst cameras for a variety of reasons that are beyond the scope of this comment.
ALL AF systems are limited by light and contrast range (just different kinds of contrast). Very low light areas or areas with no detail (e.g. flat surfaces) will lose focus. This can be worked around (AF assist lamps), however the behavior of hunting changes based on focus methodology. Phase detect hunting tends to go towards extremes (fully in or out) while contrast wobbles and steps slowly. A wide AF array should be able to see large groups of objects, but without edges (contrast) or lines (phase) there's just no detail to check. Some cameras' wide AF methodology can only check for areas inisde one point (determine a point) while others will use multiple points in the plane of focus to keep the lock. Continuous AF is best for a moving subject, this is a case where using AF-S, back-button focus, or a program/camera that can lock or hold focus is better than just sticking with continuous AF.
There's also Canon's Dual Pixel AF which is a different physical of on-sensor phase detect, but works towards the same goal. Aptina (another sensor manufacturer) also did its own on-sensor phase detect system used in the Nikon 1, but they have since cross-licensed their patents to Sony and pulled out of most of the market.
There's a basic primer on AF systems, how they work, their gotchas, and their strengths. I forewarn you, going down this path is a dangerous one. You will start attracting camera nerds who put brand above all else, and they are not for the faint of heart. However, your videos are great and if anyone can do a straightforward explanation of how all kinds of AF works, I believe you are more than up to the task.
Thanks, this is very helpful!
Wow, excellent! Best youtube comment I ever read 👍👍
I read every AF as As Fuck
"...audience watching the video might not." Hey, I'm here *specifically for* this level of detailed tech info and geekery! To paraphrase Tech Connections Guy, "spoiler: huge nerd." Thanks for the edification. 😁
And what about using two sensors for sensing 3D? We have two eyes, and that contributes to fast focusing.
Please, if you want do some research and make a very long video explaining the autofocus on old camcorders, modern cameras and mobiles phones, it would be amazing.
I second this, the history of autofocus would be both good for photography skills and super interesting
Pretty soon I'll be starting some videos on camera tech, so I'll be sure to include a video on autofocus! I'm sure there's a lot of interesting stuff to uncover
Thank you! I love your all of your videos
Also, what a nice extra S at the end! You can really tell I spend a *lot* of time with videos for this channel.
A nice S is always good in my opinion...wink, wink, nudge, nudge.
Before someone brings it up; yes there are more than just 2 airports in Chicago, but O'Hare and Midway are the major ones that you or I would be travelling through.
Technology Connections 2 both are awful to fly to or thru during the winter months. So many times diverted or stranded there.
I remember old Meigs Field for it being the default airport in Flight Simulator. Years later, I read about what happened, how it was demolished almost as a covert op, in the middle of the night and without notice. Sad.
Wow. I didn't even know that Meigs Field was gone. Guess I don't get out much, LOL.
I've always felt that MDW was a bit of a harsh landing... I like to sit near the wing and watch them "panic" to stop the plane while the engines reverse. I'm guessing the runways are a bit small. Still, living near it, it's the only one I fly out of.
SaltyHashes Midway is a square mile in size
This is part of why almost all "real" video cameras have a setting (often a dedicated button) to lock it on infinity focus. In fact, specifically in regard to a nighttime scene with "bright points of light", I've seen some camcorders that have a "Fireworks" preset which locks it on infinity focus and reduces the exposure to prevent the fireworks from getting blown out.
It's very frustrating to me how smartphone cameras are often so close to perfect, but with some serious annoyances or flaws. What I really appreciate about the camera on the OP3 is that its image stabilization is really good, even in this low light. I didn't do anything at all with this footage, and you can see during the landing that the phone was moving around pretty violently (pay attention to the window near the edges) but it coped very well. Smartphones have the benefit of so much software but sometimes implement it in strange ways.
Technology Connections 2 On my OnePlus 5T you need to tap and hold on the point you want to focus and eventually you'll get a message telling you the focus has been locked. I'm not sure if it's the same with the OP3 but I thought it would be worth a try
The autofocus lock on my Galaxy S5 never seems to stay on. Unless I'm shooting in 4k, which is overkill for dumb videos on Facebook.
You can download an app for advanced camera functionality. That’s just the default camera app.
I really wish I could just do manual focus on my S5. I'm running Android 6 and I've tried advanced camera apps but none of them work. It's especially annoying when trying to focus on something up close. I know the hardware is capable because it will be in focus for a split second before autofocus decides it'd rather make it blurry.
Ohhhh THIS is how I can subscribe to your channel twice.
I like how you find the most mundane things intriguing
I guess I just have to think about some random thing what interest me but never looked really into it, and wait for a while because this and the main channel delivered the answers every_single_time for the past few months. Stop following my thoughts, man!!! Actually, please continue.
They’re not “points” of light. They’re blobs. I suspect, as the light spreads, it leaves few actual sharp edges. It’s more like the dark and light areas are fading into each other rather than creating contrasting features that the AF system can use.
That seems like a software issue to me. When you tap the screen to focus there, it looks like the camera continues scanning that area to maintain focus. Once that area has no more contrast, the camera starts hunting. Cell phone cameras have to be dead-simple, so that people who -think vertical video is good- are novices at photography can use them. That behavior might be a compromise to get better ease of use for novices.
Also, I don't know if this is true for contrast-detection AF, but with phase-detection AF, you need detail along a line for the AF to work. Focusing an SLR on a point is difficult unless there's a ton of light.
My understanding of this problem is that when a pin point of light comes out of focus it becomes I high contrast circle with reasonable crisp edges that are easily detected by the camera making it think it is in focus. I'm sure once AI get better in camera, they will take into account single points of light, but for now they do not. Take a screen shot of a still frame when the camera is out of focus on a dot and you will see it is a nice circle, or at least in the shape of the aperture of the camera and has a high contrast edge.
That makes a lot of sense!
Agree with the assessment. Another thing is dynamic range. Contrast AF computes the sum of squared differences between neighbouring pixels (or a variation). In a dark scene, point lights are over-saturated. Suppose saturation level is S, a point light has actual level of 9S, it would still register as S. If it defocuses to a disk of 9 pixels, each pixel would still be S. The circumference increases by 3 times, and the sum of squared differences increases by roughly 3 times. Thus the AF often prefers somewhat out of focus blobs.
I had the same thought. The bokeh of a camera doesn't produce the same result as a blur, it actually makes a distinct circle on point lights. Bigger circles could actually have more high-contrast pixels than when it's in focus.
That's what I figured was happening too. A larger circle of light has more pixels around the edge. Seems like a miracle that it works at all...
If you tap and hold in the camera app you can lock the focus and exposure, I tend to do this often when filming with my phone's camera since it bounces everywhere otherwise.
Autunite Thanks man! I wish I knew this earlier
I'm going to guess that the primary reason Southwest, as a low-cost airline, uses MDW was because as they were probably able to negotiate lower landing fees compared to O'Hare and the fact that since their fleet mainly consists of 737s means that the shorter runways are adequate for their needs. In the UK, London Stansted is a similar case - it's a hub for Ryanair (an infamous low-cost airline) while most of the other major airlines use Healthrow and/or Gatwick instead
The only reason I can think of for the focus-hunting is that the MASSIVE difference in contrast confuses it. "Wait, that can't be right, I should probably refocus to make sure."
Then the few non-points-of-light in the frame also confuse it. "Huh, those points of light are clear, but these others aren't, maybe my user wants these in focus? Oh, they won't focus, because they're not actually points of light."
Try the OpenCamera app -- it has focus lock even on videos but it has no other fancy functions such as filters, video cropping, etc. Used that and never had any problem with it losing focus.
Penurious Sierra yep pros use manual focus. If you want to make pro videos dont se a mobile phone. Ps my camcorder is always set auto everything cos I'm not a pro. But when I do dabble you can get great results with manual focus.
Penurious Sierra That is one of the reasons I have it on my phone, the focus and exposure lock options. It can even rotate and crop automatically to keep the horizon level. And timelapse stuff
Another consideration might be that autofocus once "locked on" will only "fully" reevaluate the entire image periodically, but relies on pan/tilt data from the video compression chip more often to track the movement of precomputed contrast boundaries as a lazy refocus mechanism.. it could be designed to do it that way more often to reduce computational cycles... with so little variation in these scenes pan/tilt errors could add up enough it could easily get confused, lose its contrast boundaries by a pixel or two, and have to start from scratch again... if I had to design it, I’d consider that as one autofocus mechanism.
My old Sony handicam years ago had a switch for auto/manual focus. The nice part was that it also had a momentary toggle - if you pulled the switch all the way down it would focus to infinity immediately and then it would be in manual mode. If the auto focus was being screwy you could just grab this easily and tap it to inf and switch back to auto - it would usually fix the problem right away.
Delta does fly into Midway but they are the only other major airline.
I know for sure that the aircraft in the video was a Boeing 737 of some sort with the CFM International CFM56 engines. I can tell from the sound of the engines when it touches down on the *RUNWAY* . It's not the "tarmac", at least not called that, even if it's made from that material.
My understanding is that these sorts of autofocus systems require contrast on a particular axis. So they can focus on something like vertical blinds, but not horizontal lines.
While watching this, I had to keep fighting the urge to tap my screen when it went out of focus
I’ve figured it’s because with points of light and lens blur points become circles with a high contrast edge. So being out of focus optimizes the greatest area/number-of-pixels with high contrast.
PS I figure phase detection focus wouldn’t have this issue; it would have other issues, maybe lower likelihood of the focus pixel having light go to it.
Without the pilot commentary, this could be anywhere because of the darkness.
My guess would be that the camera has difficulty with depth perception, which is what focus really is, focusing on how deep(away) something is. So, with only points of light and not a continuous source of depth information, it would make sense that the camera would struggle to focus.
Could you do a series on railroad crossing devices.
I've noticed that my Sony HDR-CX760 has no problem with scenes like this, but I think there is something else going on (though I'm very likely wrong).
That's why I use Canon video cameras sometimes because external focus sensors (iAF) that everybody seems to mistake for an extra light or a stabilizer sensor.
I would imagine that the problem it's having is that the points of light that it's trying to resolve are all at different distances, and therefore it's trying to focus on a new point of light every times it focuses on another. I don't know crap about how cameras work, but it sort of makes sense to me.
After the plane touched down I felt antsy for it to get to the gate just like I was sitting on board.
Two reasons i can think of. 1 Because the sensor is dimly lit, there is quite a lot of noise that the af algorithm will struggle with. 2. A point light source with most lenses will when seriously out of focus show nicely round bokey balls which might throw off the algorithm
I hope my camera actually has a manual focus for video shooting. It just stays locked into one focus point until I zoom out and in again.
I think it does not detect contrast between pixels but tries to find straight lines in picture. The simplest (and very fast) algorithm is to use convolution function and then see at which focus setting there is greatest amount of them visible.
This is a sheer guess on my part. But I believe it may have something to do with the fact that the points of light are moving in the frame. In other words if they were static it would be in the number same pixels all the time. Yet since they are moving they could be in X number of pixels in one frame and Y number of pixels in the next leaving the system to believe that is suddenly out of focus.
It was my understanding that it only looks at a small region, first of all. Secondly, at least on a camera system I worked on years ago, that you would focus by finding a sharpness value of a frame, blur the image in the system, and get the sharpness value for that. compare that. If it is similar, it’s likely already out of focus. Then adjust the focus of the lens and repeat while adjusting that. The highest sharpness value is the proper focus, probably. Then every so often, you need to run it again to make sure you’ve maintained focus. That’s how our system worked anyway.
"ENLIGHTEN" I live for this.
Some cameras use on-sensor phase detection, which isn’t so great in low light. I know canon’s dual-pixel cmos autofocus has issues with very low light focus because it’s phase detection based.
The oneplus 3 uses a 16mp sony PDAF sensor, which stands for Phase Detection Auto Focus. Looks like that’s why you’re focus hunting.
Ooh, really? I never knew phase detection was in use in smartphones. Research fail!
BUT, given the tremendous ability of a smartphone to use software, it surprises me that there's no analysis taking place on the software side to minimize the focus-hunting. Perhaps we'll move towards that some day--best of both worlds!
Yeah I really wonder why modern cameras don’t have the ability to switch between phase detect and contrast AF when one is CLEARLY not doing the job.
Some do. This is what Sony's Hybrid AF does in their mirrorless cameras which use on-sensor phase detection, and I presume this is also available on their small sensors used in cell phones. However, it is a function of image processing, so if the AF methodology of the phone's image processor doesn't do it, then they would have to implement it.
You can force those cameras to do different focus methods based on focus modes, FWIW.
My guess is that autofocus algorithms contain an escape algorithm in case a small insect were to fly right in front of the camera.
The algorithm would start tracking the tiny insect, and should the algorithm try to always obtain maximum contrast, there is a local maximum of best contrast around that insect and the AF algorithm would not be able to focus on anything behind.
The same phenomenon would happen if trying to film anything through a window (any tiny speck of dirt on the window would steal the autofocus and never release it).
The algorithm probably detects the size of objects and ignores them if the object is too small. Making the camera's autofocus algorithm ignore isolated point lights.
Maybe it operates after a low-pass filter to avoid false triggering from sensor noise? In that case it could be possible that the points of light just don't cover a wide enough area to be significant enough for focus lockup
I noticed this too, and I think I figured it out... Maybe. Some cameras use both a laser/IR rangefinder, and as a fallback it uses the contrast detection, and they can conflict a bit. What I have found is that most phones (#ButNotAllPhones #ButSometimes) that work with alternative camera apps that have a 'lock focus to infinity' setting, which works for these types of situations. Another option is 'fireworks' mode, which should do roughly the same thing but also bring some increased brightness... I use Open Camera on Android and it works pretty good. YMMV.
But the real reason is that it looks for areas of the MOST contrast, so the tiny little points of light are probably 'too small' or considered noise to be used as valid things to be focusing on.
I would say it's because the iso is probably as high as your camera can muster there and that make for a noisy signal. So the program might read the nois around the light as right next to the lense and that would make it pull back to it. If that's not it I blame the window.
My guess would be how the lights are moving through the locations on the image sensor used for AF. Depending on which AF mode you're in.
Maybe there was something on the window? Sometimes there's a little dirt, speck, etc, and the camera tries to focus on that. Or a reflection from inside the plane.
It could also be what others have already suggested: the camera has certain point[s] that it uses for autofocus, and when there's nothing in that particular point, it struggles. Though it would seem trivial for a modern smarthphone to scan the entire image with software, especially since everything was at infinity.
I believe this happens because the camera defaults to trying to focus on a small square in the center of the screen. When you tap, you're suggesting the best place to compare contrast. And when the image changes beyond a certain threshold, it refocuses either back to the center or at the place you tapped depending on the camera app, and by then the light source has moved and it's just darkness its trying to focus on. No contrast, as you know, means poor autofocus. I don't think the software considers the bulk of the picture when focusing, just a tiny sample.
Could it be related to how out-of-focus lights look like in cameras? Differently from a gaussian blur, lights look like little disks with relatively sharp edges on camera. These edges may fool the AF algorithm, leading them to prefer the out-of-focus image to the sharpe one, since the latter has much smaller edges (although slightly sharper). The fact that bright light are probably clipping, and therefore the brightness level in the sharp image is not much higher than in the out-of-focus one, may help too.
Thank you for posing the problem. I have been going crazy with my Oppo A3S mobile jumping in and out of focus on videos and thanks to the learned kefkafloyd, I understand why. I have downloaded an app mentioned by one of the other replies and I am looking forward to seeing if it works. Love your channels. Cheers.
6:33 - 6:46 -- We need AFP95 to do the GPWS callouts during this section of video. ;)
Maybe because contrast AF works with the unfilteted, damned noisy raw image data with a lot of "fake contrast", not the heavily noise reduced processed data you seebon screen/video?
Thanks for calling ATL by its full name ;) Interesting note, unlike Chicago O'Hare, which is gerrymandered so as to be completely within the city of Chicago, Hartsfield-Jackson Atlanta International Airport has some concourses (T, A, and part of B) in College Park, GA and others in unincorporated Clayton county...oh, and a small sliver of the northern end of the airport where Delta is in Atlanta proper. This led to a tax dispute on whether the city of Atlanta (which owns and operates Hartsfield-Jackson) owed College Park back taxes over liquor sales (they do) and whether Atlanta can collect taxes on booze going forward (they can't).
It may as well be called NON AUTO FOCUS.
Do phones actually use contrast detection autofocus these days? I keep hearing about Focus Pixels and laser autofocus and different exotic techniques that supposedly replace it.
This video ruined my relationship. I was watching this video with my girlfriend, and when we saw the extra S, we both laughed really hard about it. The thing was, I couldn’t stop mentioning it. I was making references to it all night, hours after we had finished the video. Get this, I woke her up in the middle of the night to say “S.” She grabbed me by the hair and then made me sleep on the floor after I did that.
The edges of bokeh look sharper than flarey points of light.
Personally I just came back from Houston recently and took some very similar shots coming back at night. I have always used Center point focus and have never had an issue. Something to be said about using grid focusing in certain scenarios but this is probably not one of them. When I take a picture I tap my focus point so the phone knows to focus on what matters to me.
As someone who grew up in Schaumburg and always flew in and out of the Chicago area, can I just say fuck Midway? No, seriously, the one and only time I flew out of there on Southwest Airlines my bag went to fucking Hawaii when I ended up going to Phoenix. I didn't have a suit to wear for my great grandfather's funeral because of that shit.
This scenario is not easy, because the light is fuzzy because of the glow around it.
I've always understood that contrast auto focus isn't as accurate or as fast as phase detection auto focus on SLR cameras, also the auto focus region on many cameras only covers a small area in the centre of the sensor, so if the centre of the image has large dark patches auto focus may have trouble.
Does the same thing happen with other scenes that include a lot of solid color (besides darkness)? I suspect the root cause of the "hunting" is that darkness is the easiest way to flood the sensor with a single "color". Outside of scenes of darkness, fields of truly solid colors are probably fairly uncommon.
This is probably an example of "optimization for the common case".
Here's what I think is happening: your phone (OnePlus 3 isn't it?) uses Contrast Detection AF, like you said, and it needs contrast, but it needs contrast *detail* . The low light scene means that it can't see surrounding detail at all, so it defocuses a lot instead of in small steps to measure the brightest region's smallest size and stick to that focus. The thing is that the AF depends on the metering to tell it where the brightest spot is,and since that is constantly changing, the system constantly has to relocate the brightest region, and measure multiple focus stops till the smallest size is achieved. Then by the time this process is over, if the metering system tells it that there is a new brighter spot in the frame, or that the brightest spot had changed intensity, the process has to be started all over again.
The main sticking point is that there are spots of bright light in frame, but all around those, is simply featureless darkness so the region could already be in focus and this means that the entire focus range needs to be checked to compare the spot sizes.
Great ambient cabin noise
I agree completely It is baffling. My latest fireworks video is better than some I have done, but it still isn't very good. What is even more baffling is I thought you were the guy to tell me how to get it right :P
I think it boils down to the camera simply not looking for focus where the details are. And sometimes it can mistake the circle of confusion for an in focus shot mor than the actual in focus shots.
Also. Just in general. If the distance is not expected to change significantly... It's generally preferable to simply get focus and then just turn off the autofocus. That way there is no focus hunting at all.
I've had cameras do this whilst set on "manual focus" (hmmm... not that manual at all???)
Imagine a lattice of small lights in an unlit background (like nights). Imagine that you can increase or decrease the areal density of those lights. Lets say we hv a way of measuring contrast (actually a metric to indicate contrast). Now imagine two extremes- 1) with only one light. 2) very densely packed lights which are still resolvable.
First case would hv a very low contrast metric and second case would hv a high contrast metric even though for humans, it seems reasonable to term both cases being equally contrast (actually the first one might even be judged higher).
So it follows...
Depending on what autofocus technology your camera employs, it may well be trying to focus on the pane of the window glass itself. Experienced photographers usually shut off autofocus and manually focus to infinity when shooting landscapes thru windows for this very reason.
Wow never knew how noisy a plane was from inside.
Jeez... It's been a LONG time since I've seen a jet use the whole runway before taxiing off... Even at COU, which has a 6500' runway, we usually only use about 2/3 of the strip...
Midway's longest runway is 6522' BTW...
4:44 If you want to get technical, what is Chicagoland area is South East Wisconsin to North West Indiana.
I'm only a layman, but my guess would be that it's something to with your examples involving movement (and movement that is in three dimensions).
I *never* use auto-focus if I can at all help it! I long for the days where focus was achieved using MANUAL focus-ring, NOT 'up' and 'down' buttons to set it!
Well, in the plane there are two problems. Part of the basic answer to your question is lens aberrations. Exacerbating this in many situations, as here, is the fact that there's yet ANOTHER (potato quality) lens between the camera photoreceptors and your subject matter: the window. From a foot or two back, the phone might even successfully bring the window into focus. Well, not the window, but all the imperfections and dust and etc
O'hare, Chicago. Am I the only one remembering a movie starring John Candy and Steve Martin?
Note all the lights embedded in the runway.
I wonder if you'd have the same autofocus problem outside the cabin. Like, is it possible that the camera is focusing on the window due to its imperfections? I have that problem all the time with older phone cameras.
I think that's because not all the points are at the same distance. But the contrast of all the points (points are lights) is the same.
So the autofocus is trying to focus on all of the points.
but that is impossible so it will keep hunting.
But that's my guess and I have not looked into it to.
(greetings from the Netherland)
android device? you can install a different camera app to allow manual focus in video.
Are there any updated properly to support your phone... not sure - any of then allow setting manual focus in video... also not sure
Ive always been frustrated trying to take pictures at night. Only success has been with a tripod.
Whenever in an airplane or a helicopter stills and videos are random. Take a ton and toss most.
A simple (and probably partially wrong) explanation would be that the points of light don't take up enough of the image. While a naive average approach would excel at focusing in such circumstances, this is more of a rare exception to how terrible naive average approaches almost universally are for video processing. Whatever algorithm is being used probably intermittently filters the sparse points of light out as background noise or something.
I presume you have tried a sacrifice to Cthulhu, right? Goat or better?
This aviation geek is rooting for more plane videos (and maybe aviation tech!).
Also as I watched this video, I wondered what frame rate Premiere showed you? Everytime I use my Moto G5s Plus for video, I seem to get a variable frame rate (non-standard). I've always wondered why cell phones do this. My guess is probably for getting better exposed video. They have the processing power to encode and decode it quickly. Probably much more power than a normal video camera.
I assume you watch Premier 1 Driver then.
Of course!
As far as I know AF Systems Look for edges, Not points. A single point out of Focus tends to have a very sharp edge around it which confuses the systems.
Hope you had a good trip.
Thing is, the more in focus the images are, the more chance that the points of light are blown out beyond 100%. 100% white isn't very different from 100% white, whereas if it defocuses and spreads out the light, it can contrast varying shades of grey or yellow.
That was really cool. Thanks.
What do you think of the Boring Company announcement about a tunnel to O'Hare?
I have very mixed feelings about it. The Blue Line already gets you to the Loop in 40 minutes for $5 (only $2.50 headed back to O'Hare). It seems a little too much like a benefit of diminishing returns that only the relatively wealthy will enjoy. Then again, if Musk is willing to foot the bill, it seems like it can't cause much harm. Also I did hear that there could potentially be TSA screening at the station in the Loop and you would be let off inside the sterile area at O'Hare. I might even pay extra for that.
But my opinions come with the disclaimers that I am a suburbanite with a car that almost never goes into the city to begin with, so I can't put myself in the shoes of those it would benefit most. What I think bothers me most is that it may create a new class divide. I don't think it would be good for people to see the 'L' train as a second-class option, which I can't help but imagine will occur if Musk's train comes to fruition. Here in the States many have a grim view of public transit in general, which I don't want made worse.
Can I use some of this footage for a vaporwave music video?
You should have phoned ahead for an order of White Castle's! Then you could have reached out as you flew by and grabbed the sack!
It's interesting - I always thought it was only me who was infuriated by this haha... Thinking about it though with 2021 technology - this should be easy to overcome really. The answer? Machine Learning... Apple and Google are doing this all the time, so that when you take a photo of a person it is able to recognise that and adjust the camera settings automatically for the best shot. Sun in the background? No problem, auto HDR... etc. Phones also have sensors in them - accellerometers and gyros to know which direction they're pointing/how fast they are moving, etc. Surely Apple or Google could simply bake in some logic along the lines of "IF the scene is low-light AND there are multiple points of light that look like roads AND I'm moving at over 100mph, AND I'm at an altitude well above what maps is telling me the ground level is at this lat & lon, then assume I'm flying at night and keep near-infinity focus the whole time....." Something like that...
I suggest watching the end of this video on an elevator.
My guess is that the bokeh effect on punctual light sources creates a round shape with pseudo sharp edges (for example: bit.ly/2tdAf4y and bit.ly/2MNrwhE), so, under those circumstances, even when it's absolutely out of focus, the contrast detection believes it's correct. I recall some (many) years ago when some people pointed their VHS cameras to the night sky and zoomed in at stars, and they believed they were taping UFOs, because they saw "circular shapes" flying around.
noise ... and wide dark area ... not really ideal for determining the contrast variations between two focus.
maybe it's because of high iso?
Fantstic landing at night
Did I missed the point ?
graphing double slit experiment.
Oh man, I just realized how much I miss my watch beeping on the hour; it's been a couple years now that I've been a pebble user, and pebble has no piezo.
defracting grafting is the term
one thing you should not use auto focus looking thru glass.
Wow 😮
I like you better when you're not acting.
All that space to build and people choose to cram up together. I would hate to live in a busy place. I'm so lucky to be able to take a short walk and have no people or buildings around me
Tangobaldy Chicago’s reason for existing is that it is where several transportation routes cross in the middle of North America. People are clustering close to those. Yep the emptiness close by is just fine with me too.
Francesco Nicoletti reminds me of the song Telegraph Road by Dire Straights
Francesco Nicoletti It's also where the largest data center in the world is! 320 E Cermak.
Cities exist to bring people and products/services close together. I can have amazon deliver most things the same day, I have a game shop about a ten minute bus ride down the street, right next to my community college. Driving a bit farther I can get to old town where there are several nice Italian restaurants, an apple store, several clothes stores or whatever (clothes = boring), a really good build-your-own ice cream shop, and a dozen other places I’ve never been inside.
I too enjoy spending time away from the chaos and in nature, but I value the access being in a city provides as well. (We also have a mountain to hike on if one wants to be away from the city for a while, however it’s California so for most of the year it’s basically just covered in dust and dead plants, and is quite hot.)
On the other hand, as delivery gets cheaper and work moves online I can certainly see more people moving away from cities
MDW:ORD::SNA:LAX
Mobiles (cell phones in your silly country) also have sensors to know how far away the item is to focus on. As you're looking through windows that will likely not be focused (see what I did there) on clarity but strength, so extra sheets on them you won't notice, this will no doubt be picked up by the phones sensor. The beam will be hitting the glass and coming back as though you're filming something super close and explains the massive jump in focus at times.