Explaining Canon's "Pro Level AF"

[D.Sim]

I've been wondering for a while, and haven't been able to find an understandable explanation - so I'll ask here.
Can anyone explain just how the AF will work for the non-pro bodies?

What I mean is that for the AF points not working above a certain aperture - for example (from memory) - 50D outer AF points only selectable at f/2.8 and below, and the centre point at f/4 and below

So what happens if you go above those apertures? You lose the ability to autofocus? But I'm pretty sure you can still autofocus when stepped down... so just what happens?

 

[neuroanatomist]

Not sure what you're asking. All of Canon's AF points are sensitive at f/5.6. Some have additional accuracy with faster lenses (f/2.8, usually the center point), some are single lines and some are crosses. 

AF is always done with the lens wide open, regardless of your selected aperture - the lens stops down just as you take the picture.  All Canon lenses are f/5.6 or faster when wide open, and 3rd party lenses that are f/6.3 at the long end spoof the camera into thinking they're f/5.6, so AF works.  The only way to get a 'lens' slower than f/5.6 is to add a teleconverter, which costs 1 stop for 1.4x, 2 stops for 2x.

What distinguishes 'pro' (1-series) AF is more points, more cross-type points, more higher-accuracy points (f/2.8 or f/4), better tracking algorithms, greater sensitivity (better in low light), and more customizability.

There's a write-up on Canon AF here, with lots more details.

[D.Sim]

Not sure what you're asking. All of Canon's AF points are sensitive at f/5.6. Some have additional accuracy with faster lenses (f/2.8, usually the center point), some are single lines and some are crosses. 

AF is always done with the lens wide open, regardless of your selected aperture - the lens stops down just as you take the picture.  All Canon lenses are f/5.6 or faster when wide open, and 3rd party lenses that are f/6.3 at the long end spoof the camera into thinking they're f/5.6, so AF works.  The only way to get a 'lens' slower than f/5.6 is to add a teleconverter, which costs 1 stop for 1.4x, 2 stops for 2x.

What distinguishes 'pro' (1-series) AF is more points, more cross-type points, more higher-accuracy points (f/2.8 or f/4), better tracking algorithms, greater sensitivity (better in low light), and more customizability.

There's a write-up on Canon AF here, with lots more details.

Interesting... Now why did I think that? Maybe it was a misread - but for the "additional accuracy" bit, what exactly does it entail? Just how much more accurate can it be when used at those points?

I'll see if I can pull up that writeup I got the idea from - apparently its wrong...

Thanks neuro!

[BlueMixWhite]

oh i didnt know that also. Thanks Neuro.....now i understand the problem with 1Dx f8.0 issue.

[neuroanatomist]

...for the "additional accuracy" bit, what exactly does it entail?

Basically, phase detect AF is a rangefinder system, and the longer the baseline of the system, the more accurate the range estimation.  The magnitude of the increased accuracy will vary based on the subject distance - there are three points to the triangle, the further away the apex is, the less accurate the estimate (but also, the less important accuracy becomes, since DoF is deeper with more distant subjects).

In practice, that increased accuracy can also be achieved by sampling the phase data with greater frequency.  Canon actually uses that approach on the 1-series center AF point (prior to the 1D X, that is).  The 1-series bodies have several AF points that are f/2.8 in one orientation and f/5.6 in the other, but the center point is f/4 and f/8, respectively.  Since the center point is critical, it wouldn't make sense for Canon to have it be less accurate at f/4, right?  So, the line sensors for the center point have twice the pixel density of the other AF points, and that increased sampling 'recovers' the accuracy lost to the shorter baseline.

[Drizzt321]

...for the "additional accuracy" bit, what exactly does it entail?

Basically, phase detect AF is a rangefinder system, and the longer the baseline of the system, the more accurate the range estimation.  The magnitude of the increased accuracy will vary based on the subject distance - there are three points to the triangle, the further away the apex is, the less accurate the estimate (but also, the less important accuracy becomes, since DoF is deeper with more distant subjects).

In practice, that increased accuracy can also be achieved by sampling the phase data with greater frequency.  Canon actually uses that approach on the 1-series center AF point (prior to the 1D X, that is).  The 1-series bodies have several AF points that are f/2.8 in one orientation and f/5.6 in the other, but the center point is f/4 and f/8, respectively.  Since the center point is critical, it wouldn't make sense for Canon to have it be less accurate at f/4, right?  So, the line sensors for the center point have twice the pixel density of the other AF points, and that increased sampling 'recovers' the accuracy lost to the shorter baseline.

Fantastic writeup!

In there you say that Contract Detection auto-focus has some advantages (obviously some disadvantages as well) over phase-detect. Do you know of a good writeup of how modern contract detection systems work? I've read a few things here and there that it's improving pretty fast, and in a few years SLRs might start using it or something like that.

[Mt Spokane Photography]

Fantastic writeup!

In there you say that Contract Detection auto-focus has some advantages (obviously some disadvantages as well) over phase-detect. Do you know of a good writeup of how modern contract detection systems work? I've read a few things here and there that it's improving pretty fast, and in a few years SLRs might start using it or something like that.

The internet is full of good explanations about how the Canon phase detect anf how contrast detect autofocus works.  Both have been around for many many years.

Try

http://en.wikipedia.org/wiki/Auto-focus

http://www.cambridgeincolour.com/tutorials/camera-autofocus.htm

[Micke_84]

Some questions.

The 1dm4 contains cross-point sensitive af whit some lenses and 1,4X converter (ex. 70-200 f2.8 II + 1,4X) so it will be cross-point sensitive at f4 -> how do this work? why not sensitive whit all f4 lenses?

And then the 7d af, is the 7d cross-points only active whit lenses that has an aperture of f2,8 or larger?

Some questions that have been on my mind for a while!

[neuroanatomist]

The 1dm4 contains cross-point sensitive af whit some lenses and 1,4X converter (ex. 70-200 f2.8 II + 1,4X) so it will be cross-point sensitive at f4 -> how do this work? why not sensitive whit all f4 lenses?

That's a good question, and the best answer is probably 'because that's what Canon decided.'  The limitations are in the firmware.  The trend continues - for example, Canon proudly advertises that the new 1D X has five f/2.8-sensitive dual cross points, but they don't proudly advertise the fact that the 24-70mm f/2.8L, despite being an f/2.8 lens, only activates one of those five dual cross-type points.

I'm not suggesting Canon is being arbitrary or capricious here - most likely, they test these lenses with the AF system and remove firmware support for combinations that don't deliver results they find to be consistent with 1-series performance.   The lenses that are f/2.8 but only activate one of five f/2.8 dual crosses have in common that they are all 24mm or wider with an f/2.8 max aperture.  Narrower f/2.8 lenses, and faster wide lenses (24mm f/1.4) activate all 5 points.  That suggests there is a technical reason for these decisions.

And then the 7d af, is the 7d cross-points only active whit lenses that has an aperture of f2,8 or larger?

No, the 7D has 19 cross-type points with all lenses of f/5.6 or faster.  The center point is a 'dual cross' point - an f/2.8-sensitive 'x' superimposed onto the f/5.6-sensitive '+', and the f/2.8 'arms' are active with f/2.8 and faster lenses.

[Quasimodo]

Thank you all and Neuro in particular for an interesting and educational discussion here.

While some of this becomes too technical for me to follow, I thought that I will throw in my own beef with AF here. From a compositional freedom point of view, I would wish for a AF system where (cross type or not) one would be able to have AF points covering the whole area (so for instance I would be able to choose AF point in the upper right corner if I believe that would make a good composition. I asked some friends who are more versed in photography than me, and what I got out of their explanation is that because the lenses are round autofocus will tend to revolve around the center and outwards, but that a AF system (optional) where you cover all bases would not be possible because of the lenses round construction.

Does this seem right as an explanation?

[neuroanatomist]

While some of this becomes too technical for me to follow, I thought that I will throw in my own beef with AF here. From a compositional freedom point of view, I would wish for a AF system where (cross type or not) one would be able to have AF points covering the whole area (so for instance I would be able to choose AF point in the upper right corner if I believe that would make a good composition. I asked some friends who are more versed in photography than me, and what I got out of their explanation is that because the lenses are round autofocus will tend to revolve around the center and outwards, but that a AF system (optional) where you cover all bases would not be possible because of the lenses round construction.

Does this seem right as an explanation?

Not really, it's not an issue of the image circle being a circle, although the aberrations at the edge of that circle (distortion and vignetting) do play a role.

There are technical limitations on the spread of the AF points - at best, they can only occupy the middle area of the frame, because of simple geometry and optics. In a nutshell, there are four reasons for this limitation:
 

• Size of the secondary mirror. Light for AF passes through the semi-transparent part of the main mirror (most is reflected up to the viewfinder), then is reflected off the secondary mirror down to the AF sensor. There is limited space behind the main mirror, based on the necessary geometry (i.e. the main mirror has to be at a 45° angle to the incoming light, and the secondary mirror has to be behind the main mirror and at an angle of 90° to the main mirror, so it's length is limited by the distance between the main mirror and the image sensor).
• Distortion. With many lenses, the edges of the frame are subject to distortion (barrel/pincushion), and that reduces the accuracy of phase detect AF.
• Vingetting. The AF system needs a certain amount of light to work. Almost all lenses vignette to some degree, meaning there might not be enough light at the edges of the frame. For example, the EF 17-40mm f/4L has >2 stops of vignetting wide open at the wide end - that means at the edges of the frame, you're below f/5.6 and AF sensors would not have enough light to operate (i.e. in dim light you'd be below the EV sensitivity of the sensor).
• Temperature. Canon has stated that larger AF sensors are more susceptible to changes in temperature with the result that they change size, getting either larger or smaller as the temperature rises and falls. That reduces the accuracy of the AF system overall.

It's worth noting that none of these limitations apply to contrast detect AF, so using LiveView you can autofocus right out to the edge of the frame.  That's one of the 'advantages' of contrast-detect AF that I was alluding to (another is the ability to focus with a max aperture narrower than f/5.6).

[7enderbender]

 

• Size of the secondary mirror. Light for AF passes through the semi-transparent part of the main mirror (most is reflected up to the viewfinder), then is reflected off the secondary mirror down to the AF sensor. There is limited space behind the main mirror, based on the necessary geometry (i.e. the main mirror has to be at a 45° angle to the incoming light, and the secondary mirror has to be behind the main mirror and at an angle of 90° to the main mirror, so it's length is limited by the distance between the main mirror and the image sensor).
• Distortion. With many lenses, the edges of the frame are subject to distortion (barrel/pincushion), and that reduces the accuracy of phase detect AF.

Not to hijack this but I'm curious about this point. How much light are we actually losing in the viewfinder due to this setup? I'm still wondering if anyone has ever tried to change the mirror to a non-transparent one and customize, e.g., a 5DII to be better suited for full-time manual focus - while maintaining the camera's ability to meter correctly.[/list]

[Quasimodo]

Please feel free to hijack:) This site (who by the way was recommended to me by a friend of mine who works in Canon) is great, and I learn a lot from reading the different threads. I post my pictures at 500px, but I have always missed a forum where you can ask more experienced photographers about stuff. Here I have found it:)

[neuroanatomist]

Not to hijack this but I'm curious about this point. How much light are we actually losing in the viewfinder due to this setup?

Very little.  Consider - the whole main mirror is not semi-transparent, only certain areas allow some light to pass through vs. being reflected.  Yet, when you look at an evenly illuminated white wall or clear blue sky, you don't see light and dark areas corresponding to the parts of the main mirror that pass light.  So, probably less than 2-3% of the total light, and only in select areas, is 'lost'.

[EYEONE]

Not to hijack this but I'm curious about this point. How much light are we actually losing in the viewfinder due to this setup?

Very little.  Consider - the whole main mirror is not semi-transparent, only certain areas allow some light to pass through vs. being reflected.  Yet, when you look at an evenly illuminated white wall or clear blue sky, you don't see light and dark areas corresponding to the parts of the main mirror that pass light.  So, probably less than 2-3% of the total light, and only in select areas, is 'lost'.

Ya know, considering how little light actually passes through the mirror it is pretty amazing that the AF can find focus with such little light to begin with. I wonder if we are already at the limit of how semitransparent the center of the mirror can be without noticing. It seems like if they could adjust the main mirror to let more light through they could get an AF system to work in darker situations. Or am I right in assuming they have already maximized that part of the equation?