Autofocus for GEEKS: Phase vs Contrast (WARNING: Intense detail)
HTML-код
- Опубликовано: 8 сен 2024
- First 500 get a 2-month free trial to Skillshare: skl.sh/northrup14
Tony gets DEEP into how mirrorless and DSLR cameras autofocus with Contrast-detect autofocus (CDAF) and Phase detect autofocus (PDAF). You'll understand why DSLRs sometimes need micro-adjustments, why CDAF is slower (but more precise), and why CDAF cameras (like the Panasonic GH5 and S1) have focus that hunts in and out when recording video.
Music: share.epidemics...
Just when we thought Tony'd reached the apogee of photo nerdness. This was utterly fascinating. Thanks for going the extra mile to help us understand our beloved passion!
This is why i like this channel more over others, especially when Tony geeks out like this. It's a great learning experience that gives a deeper understanding of cameras work, which can only improve the way you shoot. Most channels are basically a "wash and repeat" of photography. If you want to become a knowledgeable pro photographer, this is one of the best channels to learn from because it covers a lot of aspects of being a photographer, not just about a camera and lens but everything from the obvious like lighting, models, posing etc... to computer hardware set up, website ... to boat loads of useful tips.
This is exactly the level of detail I appreciate! Very well communicated even for those of us with a shaky grasp of science. Thanks
One thing is that contrast detect AF can be improved, as companies come up with new methods of subject tracking. For example, instead of pulling a Nikon and setting the CDAF loop to a 1 second delay, some will use a delay free method and instead establish a baseline of out of focus that is allowed before it attempts to focus again, thus with minimal movement, there will be no hunting, and when a subject moves, some basic logic is used to determine if something is moving towards or away from the camera, and thus will make a smarter choice as for when to adjust. They will also keep track of peak contrast levels, in order to avoid the need for an overshoot. Some CDAF algorithms can achieve almost follow-focus type behavior, the issue is that most DSLR makers refuse to modernize their CDAF. For example, Nikon's CDAF still does not use the accelerometer data to measure camera shake and use that data along with the shutter speed to determine if camera shake is causing motion blur and then pause the CDAF loop until CDAF can be reliable used to get focus. Some smartphones will do this and it helps prevent situation where if there is a lot of camera movement that the camera does not completely lose focus or hunt endlessly. Most DSLR makers refuse to bring over some of their tracking algorithms related to motion estimation when it comes to subject tracking. This is why on a camera like the D7500, subject tracking in video always lags behind the movement of the subject. On the other hand, if using the phase detect system, even though it shuts off while the while the pellicle mirror is being moved, it can still get accurate focus of a fast moving subject, and that is because the camera estimated how fast the subject was moving prior to taking the image, and then used that data to calculate a subject distance that is likely when the sensor is finally exposed. This can be done with CDAF also, and some cameras already do this, but DSLR makers that have a separate PDAF sensor, never implement these features into their CDAF system like how the cameras using modern CDAF only systems are doing.
Coming from an Electronic technology background before going into IT, this video really helped me to understand how AutoFocus works. Thanks so much for sharing and explaining how it works. Keep on geeking!
Tony going full on nerd but this is one of the first times I saw such an easy to understand explanation of how focusing works. Kudos!
Still lost. 😄
You're always a few steps ahead from everyone else but you stay humble enough to drag us with you. Thank you!
I signed up for the free 2 months on skillshare. I’ve gotta be honest, it’s pretty amazing. Huge thank you to Tony for the free 2 month trial!
This is Amazing Tony! This is the reason I live and respect you both and your channel. Always good to hear your thoughts and ideology in photography. Keep doing what you're doing. Wish you the best health and success!
I am a technical nerd to the max, and I LOVE videos like this. Thanks Tony!
Your explanations were spot on, "in focus" :), especially on how the focus sensors are integrated into the main light sensors of mirrorless camera was an eye opener. It will make me think twice if I ever wanted to switch to a mirrorless camera in the future.
Great video, and makes it easier for people to understand why Pana have just tried to improve CDAF. Ultimately it improves image quality and accuracy, but makes it weaker for video autofocus. I'd say the S1R is probably the best studio camera along with the D850 because of the issues with phase detection on sensor.
Indeed. It's kind of ironic that Panasonic tend to be more lauded for video than stills, when their AF system is actually better for single-AF stills rather than continuous video AF. They have made great strides in improving it, and an advantage of being mirrorless is that they can use computational scene analysis in order to predict movement... but without PDAF it will always have the problem of having to pulse in and out to check for better focus. I love Panasonic cameras in every respect except for their refusal to provide PDAF. But it's nice to have a choice I guess.
I never thought I'd correct Tony effing Northrup, but, The first SLR with built in Autofocus was the Polaroid SX-70 Sonar in 1978
KoalaMeatPie yep, and it was followed by Pentax with the first 35mm SLR Camera about 3 years later with the ME-F. The Minolta 7000 actually didn’t come out for another 2 years after the Pentax.
Yes, but you would need to gently stroke the case and speak friendly words to insure success.
Dual Sharpie Autofocus Tutorial, by Tony Northrup.
Nice tutorial and I do enjoy learning how things work. Thanks Tony!
I searched up how phase detect works cos i was curious, perfect explanation video, the engineering is incredible
Keep on the good work! I think it is very stimulating to understand not only 'how' it works but also 'why'...One of the rare sites where practice and theory are dynamically working together. Excellent!
Goodness, I thought I a NERD....I quickly lost my PHASE DETECT lock on your explanation when you went in DEEP on the subject matter. HOOK EM!
Wow!! This is one of my most appreciated videos from you channel since a while!! Thanks so much for touching this subject in such an accessible way.
WOW !!! Tony explains auto-focus with math's Pythagorean Theorem , that even non-geeky nerds can fully understand to learn this important lesson..... for all camera owners !!! Thanks Scarboro
"Nerdthurp" does it again:Tony presents a topic i don't care to truly understand and it keeps me watching. That diagram also replayed a bad flashback to HS trig class and watching my G.P.A. go south.
Excellent clear explanation of a complex subject and often misunderstood. Thanks
Tony just got undone ? Are you hiding in the back somewhere Gerald?
I *really* appreciate the much-needed perspective on dual-pixel autofocus, since all other comments I've seen hail it as infallible and unmatched. As a lifetime R&D engineer, I can vouch that a huge part of design is understanding error propagation/measurement precision. So, moving the parallax of the phase detect distance measurement down to the width of one pixel seems very risky indeed. No doubt, Canon counters that some by leveraging statistics from so many pixels in aggregate. And combining that with software for subject (face) detection, should improve that (which pixels to average, which to ignore). Still, all theory needs to meet data, and your observation of focal confusion at long focal lengths says a lot. At what focal lengths does the DPAF tend to be act more reliable for you?
I keep my big Nikon 810 and 610 with the Nikon Flashes for indoor Photography because of the infrared emitter on the flash or from the special cord. You can shoot in a black room and nail focus every time. This has been true from the film days. I have gone through every other system and never matched the ability to nail focus for event photography. Aftermarket flash never work as well as the OEM flashes like the Nikon 910 and Nikon 600. I find that higher speed prime lenses really work better. My Nikon 24-110 f4. has much more accurate focus than a 24-80 with a smaller wide open F Stop.
Thanks Tony. Always interesting to hear how things work. As always, keep up the good work and keep em flying...
10 years+ of using DSLR and still blown away at just how complex and sophisticated they are.
Thanks for such a good explanation, Tony. I always wondered why people thought mirrorless systems were worse in focusing than DSLRs and now I have a proper answer with science for that. Incredible video, I am loving you nerdy stuff, one of the main reasons I have started following this channel long years ago. Good quality content delivered in understandable explanation from tech to art.
Congratulations!
I don't think that's the takeaway from this video, though. PDAF has pros and cons for mirrorless vs DSLRs, it's not that PDAF on a DSLR is always better. For instance, it can get out of sync on DSLRs and needs fine tuning, mirrorless don't need that. Also, mirrorless can do PDAF while recording video. Also mirrorless allow for autofocus to cover a larger area of the sensor. And many many other pros and cons.
Mirrorless systems tend to use hybrid AF system. In addition to PDAF on sensor itself, since his video illustrates how imprecise it can be when the angles hitting the sensors are wide apart to resolve the image, it also uses contrast detect to finalize the focusing process.
I thought I made this video already? Haha. 😜
Enjoyed the vid, Tony. Love the nerdy stuff!
I think you love nerdy stuff more than him. That's why I subscribed your channel way back!! 👍
And now we have two really good technical teachers presenting. And you each communicate in a different style. Watching both improves my overall comprehension. But you knew that, ergo, the "Haha."
Was thinking the same thing. As in you made this video Gerald.
but y'uroe Caidaann, we culndo't fluly unedrsantd waht you wree sniayg in taht aiencnt dialect ;-)
It's not funny.
You wouldn't need the hypotenuse for the two right-angled triangles you could use tan(alpha/2)=(opposite/2)/adjacent => adgacent = (opposite/2)cot(alpha/2), where the adjacent is the distance from the mid-point between the eyes to the subject, alpha is the angle subtended by the line between the eyes and the opposite is the length of the line between the eyes. Also, I think what you were trying to say is that as the phase detection sensors get closer to each other the angle that they subtend at the subject gets very small. The further away the subject, the smaller the subtended angle, so things get trickier with long lenses
That's great. I enoyed the detail. Please keep making such detailed explanations.
Thanks Tony! I've been looking for a clear and concise explanation about his for years.
You looked like you really enjoyed talking about this subject 🙂 it may take a few watches to process it for me though.
Wow that's highly detailed just as the title says! Please, please, do a video on the signal to noise ratio in mirrorless cameras, nobody talks about this..like ever
Nicely explained just right for me. If you wish I can help you revising your diffraction vs sharpening clip as this is my background
Took me a few seconds to figure out what the topic was,, but I'm so glad I clued in and pulled up a chair--great , concise explanation and stuff every (at least) every advanced amateur and hobbyist and pro should know--I know the cameras software can usually compensate for some of the distortion and vignetting we get ...Hust wondering if there are software programs that can recognize and attempt to correct banding,,,,I see it from time to time...
This is absolutely fascinating; I was actually able to follow since college physics is still fresh in my mind. Thank you for making such an informative video.
Really interesting video, thank you. I wonder if the increase in accuracy is why Panasonic religiously stick to contrast detect, despite being slower and pulsing in video? I find this interesting considering Panasonic are so videocentric.
Nicely done. My Nikon D750, I have to use liveview if I want anything really sharply in focus at even modestly open f-stops (like 5.6 or 8 on an f4 lens). Doing calibration diagrams and fine tuning sans LV, never results in focus nearly as good as the LV lockin... it's slower and necessitates using a tripod for the most part, but much of what I shoot is static and done tripod anyway - or at higher f-stop using viewfinder focus system... Another path I sometimes take is use viewfinder focus but focus stack and let software sort it out, even handheld it works out better.
I know accuracy is important, but if you rename this vid to "Contrast vs Phase - Autofocus for Geeks" it would probably perform better. CDAF/PDAF is not as recognizable I think (but I could be wrong).
@Q.E. D I have just realized by reading your comments that my mind automatically translated those terms into what they are supposed to mean. :D
OK, done, good suggestion
Great video. Pls bring more of these technical ones.
Very nice explanation. Finally I understand pdaf.
Thanks! Great stuff! You're always scientific. I appreciate that.
Thanks Tony- now I know how to go build my own camera. Watch out Sony, Canon and Nikon - we’re coming for you.
On a serious note - take a good look at the facial structure of competitive target shooters. They have a tendency to have asymmetrical eye placement. That asymmetric difference provides their brains more complex data.
Very nice video as we are used to. Small comment. The depth perception from a two eye person is made from several factors. The angel part described here is a minor one. The major is that a 3D object look different from the two eyes due to different perspective. An other minor effect is the focus created by the eye adjusting the lense.
There is a balance between knowing and the need to know. Thanks Tony for the insight. I will never buy a DSLR again (having moved to mirrorless about 4 years ago), but I admit to admiring the Nikon D850 because of its superb focus capability (said by a Sony user...)
I just love the title. Like we didn't figure on a Tony video having "intense detail". LOL What a geek, love it.
Love the camera history and the knowledge. Thanks for the link!
Any video that mentions the Law of Sines is a good video.
Werner may be right, but honestly if I see a video from Tony that has terms I don't recognize I get excited that I am going to learn something new! Another great video Tony, keep them coming!
Jesus Christ that was awesome I just wonder how much time spent studying all this.
I 💓 nerding out with Tony
I'm one of those curious nerdies and this video help me to understand. Thank you.
Great video. However, it leads to a question I've had for a while. If everything related to focusing is controlled by the camera body (and not the lens), and all the camera needs to do with the lens is tell the autofocus motor to turn right or left, why are adapted lenses less accurate when it comes to autofocusing?
I never understood this either. Not just adapted, but third party lenses too. I've just assumed it has to do with the focusing motor steps not being accurate...
I believe it's a combination of the steps on the motor and the functions to control them not being universal combined with the properties of the lenses being different than expected, for the same reason that Panasonic's DFD technology only works on Panasonic Lenses. On top of that, adapters introduce a little more connector looseness from tolerances which means the lens may droop slightly enough to send light in a different way to the sensor, for the same reason that Tilt/Shift lenses are Manual Focus only.
Each camera manufacturer uses a machine language to communicate the camera body and the lens. Adapters need to translate this language in real time to make things work between bodies and lenses, but this is far from perfect.
Some manufacturers (especially Nikon) do not like sharing the command set of their autofocus system with anyone else, so 3rd party manufacturers and adapter producers need to reverse engineer how a Nikon body communicates with the lens. Even if the body is doing the job as usual, the command sent to the lens is either not translated with all parameters of the original or the lens is losing some valuable time during the conversion.
I gotta say this because no one else has and it's important: that is a beautiful shirt on you, Tony.
Lol thanks it's new
It seems stacked sensors are the answer to banding and rolling shutter with phase detect in mirrorless as Sony have shown with the A9. Interesting though its reduced to 5fps in mechanical much like the EOS R. Probably why the A7iii has the half stop advantage at half the fps speed. Stacked sensor=more speed, no rolling shutter/reduced banding but reduced light.
The point about phase detect and telezooms seems to have eluded the A7iii/A9. Adapted lenses like the Sigma 150-600c work faster than I have ever seen on Dslr with its MC-11 counterpart with the most recent firmware updates. I think one aspect overlooked in the video about phase detect is the amount of AF points that are available on mirrorless sensors compared to Dslr which are surely going to provide huge improvements to telezoom accuracy.
Dang Tony! I'm a geek but you are turning me into a camera geek.
Gary Friedman has some great videos demonstrating AF capabilities and limitations, including how non-cross-type AF points aren't able to pick up horizontal lines. Thanks to your phase detect explanation, it totally makes sense why this is the case.
the main difference is phase auto focus represent open control system where there is no feedback to know the result while in contrast auto focus which represent closed loop control system there is feedback and continuous action to get best result.
Kudos Tony! Now so much more makes sense about new dslr vs mirrorless vs old school!
Great stuff! Now i finally understand the difference between CDAF and PDAF. Thank you.
Thank you for the math lesson. I finally understand how our eyes measure distance.
I think this is my problem, with my D800+35art. using live view its tact sharp, using OVF its backfocus. can you give your insight? TIA...
Absolutely fascinating Tony, thanks!
I would love to see a more in depth video about just phase detect AF on current mirrorless cameras, without the discussion about DSLRs taking up a large part of the video. What changes have been made in the last four years?
Tony, you are one smart dude! I love these videos!
I just love those classic, "nerdy" Tony Northrup videos
I like the youtube-style sitting at the computer jump cut deluge of info style!
Always appreciate these videos. The more you know about the physics of photography, the better you will understand how to spend your money and why things cost certain prices. Cool stuff ;)
A very educational video
Fascinating Tony. Thank you
The best info for a geek, comes from another geek. Thanks Tony
I remember this was this kid with one eye in class, and at recess we would all take turns running up to him and punching him in the face, because he could never hit us back because he would always miss. XD I'm not proud of it, but it was really funny still to this day!
Wow! So much new information. Nice job!
That was intensely detailed on the details of autofocus
Excellent. Very nicely done.
Cool -- thank you for explaining that in 15 minutes -- not a waste of my time at all!
Your latest videos are looking so good. I’m loving the lighting
Wow. Great video. Spot on, Tony. Can I have a bit more about dual pixel and Sony Eye detection with examples? That was really interesting and I feel part 2 would be in order.
Canon has a couple of good videos about how DPAF works, but really, I provided most of the public detail. I think the algorithms do most of the work and they're secret. Re: Sony Eye detection AF, it doesn't seem to be related to the focusing at all; it just helps select the right focusing point, and then standard PDAF focusing takes over.
Great little lecture on modern AF. Fascinating. Thank Tony!
Great explanation, thanks much!
Well for me I found out it really doesn't matter as long as the focus is quick and accurate. I switched from a nikon d200 and d300s to a fuji xe1 for the weight savings and somehow the fuji xe1 is actually slightly better than my nikon dslrs were to focus in both time it took and ability. That makes no sense to me as I thought phase detect was supposed to be miles faster but the xe1 is contrast detect only and im not having any problems aside from trying to photograph something that is all black or all white but my nikons had issues with that too. I mean I figured with the xe1 I'd have to be in manual focus most of the time which was fine with me but no im not having to at all. I just find it really interesting especially when thinking of contrast detect I usually think of how long it used to take dslrs to focus in live view. Maybe its something to do with mirrorless cameras being mirrorless idk but it definitely seems like more research into the actual speeds and accuracy of one vs the other in real world situations is warranted.
Great video, this is why you're the best in RUclips
Tony, are you suggesting the banding in the EOS R and Nikon Z6/Z7 is a result of these physical artifacts on the sensor. I had thought that in the recent firmware update on the Canon that the issue had been solved and now I am a bit confused as if this is a physical issue, then the only way there can have been a solution is for the camera to make up some replacement data for where the bands are as opposed to actually fixing the issue. Also in relation to the issues with mirrorless cameras and telephoto focus performance, is the issue that the camera will indicate that focus has been achieved when it has not, or that the camera will not find a focus when you press the trigger? By the way, this is a great video thanks for all the thought that goes in to the content from your channel!
Wow such dirty mirror in that D850 Tony :D
Nicely explained in that video, I'm sure everyone can now understand the differences.
Had no idea that camera sensors acted this way. Makes me more interested in getting a new camera now.
Wow. Thought I understood it before --- wonderful exposition of the details, thanks.
incredible technology. I wonder how phone cameras implement PDAF, probably using the 2 cameras?
Educational and nicely done.. Thank you..
Fantastic. I enjoyed all the nerdy explanations
Good stuff! Keep 'em coming.
What about hybrid autofocus? Would be nice to hear about that.
Hybrid basically just uses PDAF to get an estimate of the distance (remember it's not 100% accurate because it's still using photosites) and then once the lens reaches that calculated distance it quickly uses CDAF to make sure it's 100% sharp. The reason for this is because the integrated phase detect sensors on mirrorless cameras are actually less accurate than dedicated phase detect sensors.
Outstanding video
Great video as always. Thank you
Yes more nerdy videos please
A great video Tony. Makes so much sense now as to why my Canon 60D can sometimes have a difficult time getting really sharp focus. I keep wondering what I am doing wrong but the truth is, there are situations when the camera just isn't going to give me sharp focus. Thinking back about some photos I've taken, I do get better results with a prime lens. I can finally forgive myself and blame the camera, hahaha!
That was good. One thing to note is that mirrorless cameras with PDAF combine it with the CDAF mode, so as it gets close, then the latter can do the find tuning.
On a historical note, prior to AF there were split screen focussing systems and the theory by which they work is related to the way that PDAF works. Namely two images with a phase difference that line up where the image is in focus from two different paths through the lens, but using a prism system on the viewing screen that would show displaced portions of the image in two halves when not in focus. Split screens are not something you ever see on modern DSLRs.
nb.I bought one of those original Minolta 7000 SLRs (as it was called in the UK) and am amused to see that it was a British advertisement, not an American one where I guess they were called Maxxums. It's also worth noting that Minolta were found to have infringed on patents owned by Honeywell and had to pay that company $96m in damages in 1992, a fair amount in those days.
*some* mirrorless cameras use hybrid focusing - true - but the latest Canon R cameras, and I think newer Sony E cameras, completely forego CDAF and rely only on PDAF.
@@TonyAndChelsea The A7 III and A6400 both have CDAF points, and they are pretty new. Besides that, as contrast AF is just a software feature using standard pixels, it's going to be difficult to say for sure that it isn't being used unless the company explicitly states it uses hybrid AF.
@@TonyAndChelsea Having replied with some earlier, I have gone through the DPReview camera comparison tool and it reports contrast AF (in addition to PDAF) on all the models I checked, which at the Sony A7 iii, A7 ii, A7R iii, A7S ii. A6500, A6400, Canon EOS R and EOS RP, the Nikon Z6 and Z7.
I haven't bothered with looking at the other manufacturers. As I understand it from other sites, the contrast AF is used to finally nail focus as confirmatory process and doesn't involve the hunting process, at least not visibly. The most modern lenses can perform absolutely tiny fine movements which might not even be apparent to us unless magnified.
My first digital camera, the Sony DSC-V3 had laser autofocus and it was glorious.
I came from 80d to EOS R. I shooting pics with this camera only half year but i can say responsibly, mirrorless canon has much better autofocus. I still shooting with the old behavior and now from all events I usually have 30-40 % more pics in focus every time. So I still have some problems with more pics then I need to send to my clients))
Wow. Geeky but quite enjoyable.
😊👏🏼👏🏼👏🏼
"The law of since would allow you to math the distanse (...)". I just love this sentence! 🥰 (4:38)