Canon’s EOS-1D X Mark II equivalent mirrorless is coming sooner than originally thought [CR1]

Apr 25, 2011
2,502
1,878
I am interested in the general theory about on-sensor AFMA and its limitations, and how it will affect my photography. Here are some good articles of how PDAF actually works and on-sensor limitations: http://www.clarkvision.com/articles/understanding.autofocus/ and https://photo.stackexchange.com/que...ection-autofocus-need-adjustment-for-accuracy
On, I see why these "explanations" confuse you. They totally omit the description of how PDAF actually works.

(Maybe I should write one myself, but it would take a couple of hours of my life, so I'd rather avoid it)

PD uses phases from opposite sides of the lens, and these may be affected by aberrations of the lens and differences in path length because of mechanical errors.
They are affected in exactly the same way as the focus is affected.

In Canon's DPAF, it is possible (and common) that pixel pairs receive non-equal exposure even when they are in focus (the main reason for it is that the EF lenses are not fully image-space telecentric, but tilted or shifted optical axis may also play a very small role). However, if the lens is in focus, there is no phase shift in the dual-pixel array, no matter where the lens optical axis is.

On-sensor PDAF requires contrast detection or its equivalent for final accuracy:
Only when the PDAF points are sparse. Not Canon's case.

Read this to see how AFMA can still be necessary if you are not convinced: https://eduardolibby.com/2018/12/22/nikons-z7-requires-af-fine-tuning/ ("The claims that on-sensor phase-detection autofocus sensors in Nikon’s Z7 does away with autofocus fine-tuning are wrong. I had to use AF Fine Tuning to get my lenses to focus properly.")
AFMA may be necessary for a multitude of reasons, but why do you think that adapter-introduced tilt is one of them?
 
  • Like
Reactions: 1 user
Upvote 0
May 11, 2017
1,365
635
What we really need more than anything is 35/2 and 50/1.8 that stick out no further than the grip, so we can have our R's in our backpack at all times. And with f/4 trinity you could see quite a number sold to even the reporters who today have no interest in f/1.8 given that their limited-range trinity is f/2.8.

My guess is that it may be quite a while before there is a canon lens that you can put on an R camera that smaller than the RF 35mm F1.8. I don't know how small you can make a 35mm lens for the RF mount, but there has to be a practical limit.
 
  • Like
Reactions: 1 user
Upvote 0
Nov 1, 2012
1,549
269
My key item why I shy away from mirrorless is tracking accuracy in dim gyms. That's my reason for 1DX. Everything else should be there more or less already:
-enough fps (Canon can do it, the question is if they will do it)
-generic good all-around camera

Two items I'm not sure if/how mirrorless will compare to 1DX
-comfortable to hold with 70-200 for 8 hours. Maybe with grip it'll be good?
-8 hours of shooting with 2 batteries, ~4000-6000 clicks. Grip might help but probably still not enough? Might need 4 batteries? 1DX can do that ~with one battery, although I carry second one just in case.

Items I don't care:
-1 card slot vs 2 card slots. 2 is nice but not mandatory for my use.

Nice to have items:
-240fps video at 1080p (480p even better)
 
Upvote 0

AlanF

Desperately seeking birds
CR Pro
Aug 16, 2012
12,299
22,371
On, I see why these "explanations" confuse you. They totally omit the description of how PDAF actually works.

(Maybe I should write one myself, but it would take a couple of hours of my life, so I'd rather avoid it)


They are affected in exactly the same way as the focus is affected.

In Canon's DPAF, it is possible (and common) that pixel pairs receive non-equal exposure even when they are in focus (the main reason for it is that the EF lenses are not fully image-space telecentric, but tilted or shifted optical axis may also play a very small role). However, if the lens is in focus, there is no phase shift in the dual-pixel array, no matter where the lens optical axis is.


Only when the PDAF points are sparse. Not Canon's case.


AFMA may be necessary for a multitude of reasons, but why do you think that adapter-introduced tilt is one of them?
I'd be most grateful, and I am sure sure many CR members would be also, if you did take out the time to write something. I am here to learn and to pass on what I have learned, and I am eager to hear from you.
 
Upvote 0

Ozarker

Love, joy, and peace to all of good will.
CR Pro
Jan 28, 2015
5,933
4,336
The Ozarks
Prediction: Canon will release an awesome camera. The perpetually negative, who'll never buy a camera at that level no matter what, will %itch and moan and tell us why they will switch to Sony if Canon doesn't "get it right" in 3-4 years with the 1RX II. I'll recycle this comment at that time. Oh! And we'll hear about all their friends at the local camera club switching too. Meanwhile, Sony's market share will slide to 8%. We'll also keep hearing about how an empty tube (adapter) destroys IQ and is a potential point of failure, and complaints it doesn't come in white to match their big tele lenses. We'll get links to all the articles and YouTube videos they think prove their points. In the mean time, somehow people will still be taking great photos and videos with piss poor Canon gear. Oh! And all the design experts / engineers who'll fake knowledge they gained thanks to google will be there too.
 
Last edited:
  • Like
Reactions: 2 users
Upvote 0
May 11, 2017
1,365
635
Prediction: Canon will release an awesome camera. The perpetually negative, who'll never buy a camera at that level no matter what, will %itch and moan and tell us why they will switch to Sony if Canon doesn't "get it right" in 3-4 years with the 1RX II. I'll recycle this comment at that time. Oh! And we'll hear about all their friends at the local camera club switching too. Meanwhile, Sony's market share will slide to 8%. We'll also keep hearing about how an empty tube (adapter) destroys IQ and is a potential point of failure, and complaints it doesn't come in white to match their big tele lenses. We'll get links to all the articles and YouTube videos they think prove their points. In the mean time, somehow people will still be taking great photos and videos with piss poor Canon gear. Oh! And all the design experts / engineers who'll fake knowledge they gained thanks to google will be there too.
Either that or it will all gradually fizzle out because nobody cares which cameras are best because it becomes more and more clear that all the cameras are pretty good and all the magic numbers don't mean very much unless you are trying to photograph bats flying around inside a dark cave. And the market shares will sort themselves out without much help from the internet, as they always do.
 
  • Like
  • Haha
Reactions: 5 users
Upvote 0
That only makes sense if the adapters contain optical elements, which these don't. Adapters do increase the chance of there being tilt between the optical axis of the lens and the sensor plane (due to slop in the interface), but the chance of this having any significance in the real world is minimal (unless your real world involves shooting flat planes perfectly aligned to your sensor).
When I am speaking of surfaces I am speaking of the mount surfaces not glass.
 
Upvote 0
A tilt induced by an adapter will affect the phase detect AF, and would require AFMA even for mirrorless (contrast-based AF is relatively less affected).

Tilt affects the plane of focus, so of course AF is affected, but I'm not sure how AFMA comes into play, or why CDAF should be different from PDAF? Unless you're suggesting that AFMA can compensate for tilt?
 
Upvote 0
Apr 25, 2011
2,502
1,878
I'd be most grateful, and I am sure sure many CR members would be also, if you did take out the time to write something. I am here to learn and to pass on what I have learned, and I am eager to hear from you.
OK.

After long thoughts and some ugly and overcomplicated drawing attempts... let's start from this simple picture from Wikipedia:

Cirles_of_confusion_lens_diagram.png


The middle row shows a point object (on the left) in the plane of focus of the lens perfectly focused on the sensor (on the right).

The upper and the lower rows show how the same lens focused at the same distance will render point objects in front of and behind the plane of focus correspondingly.

The circle that out-of-focus point object produces on the sensor is called "circle of confusion" (or "circle of indistinction"). If it's intentionally made big enough, it is known as a "bokeh ball".

When it comes to phase-detect autofocus (and DPAF in particular), two things about the circle of confusion are important:

1. For the object in front of the plane of focus, left half of the circle comes from the left side of the lens, and the right half of the circle comes from the right side of the lens; for the object behind the plane of focus, it's inverse.

2. The size of the circle of confusion is a monotonous function from the amount of defocus. When the object is in focus, the circle of confusion is zero (this is true only for the out-of-focus circles, but the other kinds of blurs practically don't depend on from which side of the lens the light comes).

Now, what if we are able to split the light coming from the left and from the right half of the lens and to project these halves on the different sensors? Then one sensor will see only one half of the circle, another one will be able to see only the other half, and by measuring the offset ("phase shift") between the geometric centers (or some similar features) of the halves we can predict the direction and the amount of defocus.

Of course, typically we don't have a point object we need to focus on. But if we have some local contrast on or at the border of our out-of-focus object, it will be blurred but hopefully still recognized by our autofocus system, and we will be able to calculate its phase shift.

In the case of DPAF, there are lots of things that prevent us from clearly separating the rays from two halves of the lens, so we won't be getting the exact halves, but that mostly affects the amplitude difference and not the phase shift. And again, if we are in focus, the phase shift is zero.

In the case of a dedicated autofocus sensor, we can precisely extract the particular areas of the lenses from which we want to calculate the phase shift. It makes autofocus more precise and less computationally intensive, but... it will focus the lens exactly on the dedicated autofocus sensor, but not necessarily on the image capture sensor.
 
  • Like
Reactions: 1 users
Upvote 0
Mar 2, 2012
3,186
541
I'd be most grateful, and I am sure sure many CR members would be also, if you did take out the time to write something. I am here to learn and to pass on what I have learned, and I am eager to hear from you.

I suggest reading neuroanatomist’s article here:
 
  • Like
Reactions: 1 users
Upvote 0
Dec 13, 2010
4,932
1,608
Sadly, I think that whatever Canon comes out with will be so severely trashed by the “media” in the industry that it doesn’t matter how good it is. I know people in a few of the biggest shops around in my country and they’ve all said “we don’t sell any Canon nowadays, it’s just Fuji, Sony and Nikon”. They all have the RF85 in stock which they shouldn’t. They also had the 28-70 and 50 in stock, super special offers on the R and I think it’s taking a toll on Canon what the trolls of YouTube and comment sections have done that past few years.

It’s not true what they claim, but people looking for a camera sees it as truth, because it’s everywhere.
 
Upvote 0
Feb 7, 2019
411
478
UK
Sony’s specs have exceeded Canon’s for years...over which time Sony failed to capture ILC market share from Canon. But hey, maybe the next time Newton lets an apple go, it will float up.

Canon are like Liverpool football club, full of history, one of the best club teams in the world, dominated in the past, loved by millions.

Sony are Manchester City. Probably the best club team in the world. Used to be rubbish, only got where they are now because of vast investment, boring.

Man City won the premier league last season. Nobody cared. Everybody expected it.

Liverpool has FAR more supporters globally.

That’s how I feel about Canon and Sony anyway! If you’re from the UK you might get it. If not, then I apologise! :D:D
 
Upvote 0

AlanF

Desperately seeking birds
CR Pro
Aug 16, 2012
12,299
22,371
OK.

After long thoughts and some ugly and overcomplicated drawing attempts... let's start from this simple picture from Wikipedia:

Cirles_of_confusion_lens_diagram.png


The middle row shows a point object (on the left) in the plane of focus of the lens perfectly focused on the sensor (on the right).

The upper and the lower rows show how the same lens focused at the same distance will render point objects in front of and behind the plane of focus correspondingly.

The circle that out-of-focus point object produces on the sensor is called "circle of confusion" (or "circle of indistinction"). If it's intentionally made big enough, it is known as a "bokeh ball".

When it comes to phase-detect autofocus (and DPAF in particular), two things about the circle of confusion are important:

1. For the object in front of the plane of focus, left half of the circle comes from the left side of the lens, and the right half of the circle comes from the right side of the lens; for the object behind the plane of focus, it's inverse.

2. The size of the circle of confusion is a monotonous function from the amount of defocus. When the object is in focus, the circle of confusion is zero (this is true only for the out-of-focus circles, but the other kinds of blurs practically don't depend on from which side of the lens the light comes).

Now, what if we are able to split the light coming from the left and from the right half of the lens and to project these halves on the different sensors? Then one sensor will see only one half of the circle, another one will be able to see only the other half, and by measuring the offset ("phase shift") between the geometric centers (or some similar features) of the halves we can predict the direction and the amount of defocus.

Of course, typically we don't have a point object we need to focus on. But if we have some local contrast on or at the border of our out-of-focus object, it will be blurred but hopefully still recognized by our autofocus system, and we will be able to calculate its phase shift.

In the case of DPAF, there are lots of things that prevent us from clearly separating the rays from two halves of the lens, so we won't be getting the exact halves, but that mostly affects the amplitude difference and not the phase shift. And again, if we are in focus, the phase shift is zero.

In the case of a dedicated autofocus sensor, we can precisely extract the particular areas of the lenses from which we want to calculate the phase shift. It makes autofocus more precise and less computationally intensive, but... it will focus the lens exactly on the dedicated autofocus sensor, but not necessarily on the image capture sensor.
Thank you for that very clear explanation, which is much appreciated. The two articles I quoted were basically correct but your exposition is far better and gets to the crucial points.

So what are the main reasons for requiring AFMA for on-sensor PDAF?
Canon's DPAF does combine characteristics of CD with PD, which is advantageous for accurate AF. Does that and the density of AF points mean that it's going to be slower than rivals with fewer PD detectors combined with CD?
 
Last edited:
Upvote 0
Sadly, I think that whatever Canon comes out with will be so severely trashed by the “media” in the industry that it doesn’t matter how good it is. I know people in a few of the biggest shops around in my country and they’ve all said “we don’t sell any Canon nowadays, it’s just Fuji, Sony and Nikon”. They all have the RF85 in stock which they shouldn’t. They also had the 28-70 and 50 in stock, super special offers on the R and I think it’s taking a toll on Canon what the trolls of YouTube and comment sections have done that past few years.

It’s not true what they claim, but people looking for a camera sees it as truth, because it’s everywhere.
No doubt there are a few people who make a lot of noise and who are continually deriding Canon, but Canon still sell more camera equipment than Sony or Nikon, and a lot more than Fuji. Maybe those of us who still use a Canon camera are just not interested in joining the debate and we are much happier to go out and take some photos instead.
A couple of weeks ago I attended an event that was sponsored by Sony. As I walked through the door I was mobbed by a crowd of Sony zealots who wanted to know why I had not yet "upgraded" my Canon 5D mk4 to a Sony a73 or A9. They each had their own story to tell about how they once used a Canon or Nikon DSLR but they had now seen the light and moved to Sony. Well good luck to them, but in the practical session I did not notice that their pictures were any better than those that were taken by the few Canon and Nikon photographers who were present.
 
  • Like
Reactions: 3 users
Upvote 0

Del Paso

M3 Singlestroke
CR Pro
Aug 9, 2018
3,276
4,158
Prediction: Canon will release an awesome camera. The perpetually negative, who'll never buy a camera at that level no matter what, will %itch and moan and tell us why they will switch to Sony if Canon doesn't "get it right" in 3-4 years with the 1RX II. I'll recycle this comment at that time. Oh! And we'll hear about all their friends at the local camera club switching too. Meanwhile, Sony's market share will slide to 8%. We'll also keep hearing about how an empty tube (adapter) destroys IQ and is a potential point of failure, and complaints it doesn't come in white to match their big tele lenses. We'll get links to all the articles and YouTube videos they think prove their points. In the mean time, somehow people will still be taking great photos and videos with piss poor Canon gear. Oh! And all the design experts / engineers who'll fake knowledge they gained thanks to google will be there too.
This is indeed to be expected, and, as usual, trolling will come from the same Wonderful Sony Fanboys, never from Nikon or Fuji.
But you are right: the camera will be awesome!
 
  • Like
Reactions: 1 user
Upvote 0
Apr 25, 2011
2,502
1,878
So what are the main reasons for requiring AFMA for on-sensor PDAF?
I think it's because some lenses cannot be focused both fast and accurately at the same time, and AFMA helps to avoid focus hunt.

Canon's DPAF does combine characteristics of CD with PD, which is advantageous for accurate AF. Does that and the density of AF points mean that it's going to be slower than rivals with fewer PD detectors combined with CD detect?
It is more computationally intensive, which doesn't mean that focusing by itself is slower, but it may lead to slower frame rate with continuous AF. And of course all the drawbacks of the need to deal with 2x more pixels than the competitors have for the sensors of similar resolution.
 
  • Like
Reactions: 1 user
Upvote 0