Autofocus

[bfitzgerald]

Just a general chat about AF and A mount.

With the new A33/55 models (brief hands on with an A55) Sony have changed the AF array to 15 points and are now using 3 cross type sensors. When I tried the A55 in a shop the AF was indeed fast (centre point) However moving moving to the outer points (the non cross type ones) again the usual struggling to lock (low light esp)
I've had this on all A mount bodies issues with the single line sensors (seem ok decent ish light/contrast) pretty iffy at other times. Question is why don't Sony add more cross type sensors to their system?

And still using the film 7 to this day..makes me wonder why the famed "double cross" f2.8 sensitive does not make an appearance even on lower end bodies (been 10 years right?)

Moving onto previous bodies the Km5d was never a demon for AF..but why was that? Strange to me as previous film bodies were pretty decent and speedy for AF. Film 5 is quite decent (usual issues for the outer sensors low light etc as above) Not much in it between the 5 and 60 hard to tell much, using both the 5d and 60 (a lot) I remain puzzled why some still to this day suggest they are the same AF wise, they're not! Though neither is as good as the film 7 for tracking and accuracy.

On the A200 I noticed the AF spec went from EV-1 to EV0 (I noticed not sure anyone else did) And I have to be honest whilst the A200 is indeed quite a bit faster v the 5d AF wise it's not very good in low light accuracy is weak (I'm actually scared to use it without a flash giving AF assist) Seems recently the newer bodies are back down to EV-1 which is a good sign.

Still wondering why we don't have the double cross of the 7 in more bodies (I think the FF ones do have this)

I'll reserve judgement on the Pentax until I've more lenses to look at the AF performance, but so far I'd say it's x3 the speed of the 5d and certainly faster than the A200. I also notice (like the film 7) lots of AF micro adjustments going on with it. The 5d was the exact reverse slow to respond to minor changes in distance though I admit I'm not much of an action shooter thus less demanding than some.

And there seems to be a consensus that the A700 didn't match or exceed the AF of the film 7?

Any thoughts about AF in general. Is it time to put more cross sensors in even on lower bodies? Why don't we have the f2.8 sensor in more of them? Is the processor performance of the AF system a cost issue?

[David Kilpatrick]

The main reason, I think, is cost - though it's hard to see how it can be all that significant in the overall price. Cross sensors can only work for 'outer region' sensor positions either with very good lenses, or with limited accuracy. Most lenses begin to suffer from aberrations which affect the focus point of radial/sagittal versus tangential/meridional frequency (I use both pairs of terms here because some MTF charts use one terminology, others use the other, they both mean the same).

Dead centre, a cross sensor or an f/2.8 sensor has the best chance of a single true focus point applying to both axes. That's a good formula for positive lock-on quickly. Off centre, f/2.8 sensors are likely to shift the focus to match field curvature and cross types are likely to find two possible focus points, one applying to radial, the other to tangential resolution. With a live view camera and 14X magnification you can see this actually happen when you fine-focus a lens and observe the image nearer the edge of the frame.

The AF module in the A55/33 benefits from a much simpler mirror geometry before the light hits it. Just the single surface, which is full size. The problem with the bottom-mounted AF modules (in the 'footwell' of the camera) is that they use a small central area of the main mirror, then a second mirror, then a THIRD mirror which folds the light back to the sensor lenses. This folded design is used to avoid having an extra deep 'base' for the camera body.

The A33/55 don't have any such folded design, or any smaller mirrors to restrict the optical geometry. The rays go directly from the trans-mirror to the lenses of the focus array. This design is less dependent on individual lens design for efficiency, and I would guess that it's probably using f/5.6 sensors like Nikon with some latitude for smaller apertures at longer focal lengths - not the f/7.1 sensors used in the earlier Minolta and Sony modules.

These 'apertures' are often misunderstood. You have to think of an f/5.6 sensor being like the rangefinder on a Leica M, and an f/7.1 sensor being like the rangefinder on the Leica CL. It's not really to do with light, or aperture, it is like a rangefinder base difference. The M has the two windows further apart, so its rangefinder is more accurate than the CL.

It's also possible that the overall simplification of geometry and alignment has enabled off-centre cross sensors to be efficient and practical. The line sensors in the extreme edge of the field are rotated to cross the radial resolution, so they never respond to the 'other' axis of aberrations. That means they never focus at the ideal compromise, always preferring slight back focus, but focusing on the highest contrast 'direction' of resolved detail taking into account astigmatism, coma and lateral chromatic aberration.

I don't know how Canon/Nikon manage to make cross sensors work across an entire large area of the image. I would guess they are using much more complex data, from both axes of the cross, and locking the lens at an averaged position - but they already do this across multiple sensors as well.

One thing is certain, the simpler geometry of the A55/33 has paid off in a much more accurate and faster phase detect AF system without apparent extra costs.

David

[agorabasta]

These 'apertures' are often misunderstood. You have to think of an f/5.6 sensor being like the rangefinder on a Leica M, and an f/7.1 sensor being like the rangefinder on the Leica CL. It's not really to do with light, or aperture, it is like a rangefinder base difference. The M has the two windows further apart, so its rangefinder is more accurate than the CL.

Quite right!

Those AF 'apertures' are most deeply misunderstood by the vast majority. In fact, the PDAF has just a window of phase shifts that it's better tuned to work with. The greater the 'aperture', the better the AF performance is with the larger phase shifts that occur with the larger lens max apertures AND/OR with the longer focal lengths (because the DOF relative to focal length goes down thus increasing the phase shift).
Then if PDAF is used with the UW lenses, the issue to struggle against is the phase shifts too small (the DOF too large), and then we need very small 'aperture' AF sensors. So focusing the UW needs AF sensors capable of doing very small 'apertures' like f/22

And those 'phase shifts' I talk about here are a very simple thing - essentially the same as blur radius for a point light source. Different light paths cause a blurred spot image of a point source on a main sensor, or they cause two smaller spots shifted vs each other in the AF rangefinder.

And also quite obviously a hybrid PDAF/CDAF of sorts is a natural development to happen eventually to the PDAF systems - those 'lines/crosses' may get substituted with 'squares/rounds' containing the former, and those 'squares/rounds' may get the ability to CDAF within their areas as processing power permits.

[agorabasta]

A sidenote on UW AF - as the matters stand now, no PDAF can serve the UW lenses well enough; so using an a580 in CDAF mode with UW lenses seems to make tons of sense...

Wonder how the new Sigma 8-16 behaves in such a setup?

[David Kilpatrick]

Sadly, the 8-16mm doesn't focus well in CDAF. It focuses too fast to refine the adjustment, and even when it does lock focus, it won't confirm a positive result. The Sigma 70-300mm doesn't CDAF reliably either, though it will sometimes work OK. The 70-300mm SSM provides a direct comparison, and the operation is very different, with a positive green light/beep focus confirmation. Their behaviour on the NEX-5 with LA-EA1 is exactly the same.

David

[agorabasta]

Thanks, David.

I guess it's 'not this time' for Sigma. But I'll be watching them closely. I think they have every chance to get it right pretty soon, but things happen... like 'snatching defeat from the claws of success'

[twm47099]

The 7 had the best AF of all the Minolta cameras. When KM took over and released the Maxxum 70 (the name was an insult to the 7 class, at least it was called the DYNAX 60 in Europe), the AF was significantly worse, except that they added the ability to switch AF sensors when tracking moving subjects and showed which sensor was currently tracking. (The earlier cameras such as the 7 would track with different sensors, but the sensor indicator in the VF didn't light up in AF-C.)

So they added a feature and took away others (including the double cross).

I think the KM7D used essentially the same AF as the 70 (may have used a different size sensor for APS-C, but it behaved the same and no double cross.

The A700 added the double cross back, but instead of the f/2.8 sensor being a "X" it was just a horizontal line sensor with higher accuracy if it could lock (and it requires a near vertical to lock on.)

I know a lot of people like to say you can't shoot macro with AF. But with my 7 my handholding technique was to use the Minolta 100 f/2.8 macro (I have the original version) and Minolta ring flash, set the camera to AF-C with the center sensor active, and compose the shot. The camera would lock focus almost instantly and I would slowly rock back and forth toward the subject. By letting myself oscillate towards and away from the subject I controlled both shake and sideways motion. The camera tracked my motion perfectly. When I liked the composition, I released the shutter, and the flash stopped all motion.

I've tried with the 7 and a line sensor at a rule of 3rds intersection - it sometimes works, but isn't as reliable as the center. I've tried the same technique with my 700si (no cross sensor) and it doesn't work. It also doesn't work with the 7D - the camera loses track as my direction changes. The A700 almost works. Sometimes it tracks well, other times it stops and has to restart.

It was really a shame that the 7's AF technology was "lost" rather than being improved upon. Hopefully, the AF algorithms that Sony has developed and demonstrated in the A850/900 combined with a new sensor module will be a step towards lifting them to be number 1 in AF.

tom

[bfitzgerald]

I can't agree the 60's AF is significantly worse..it's not as good as the 7 but it's far from as slow as the digital bodies that arrived later.
I've used the camera extensively for years..it's a complete mystery why this keeps getting brought up. Essentially the 5 and 60 are very close in AF performance very hard to tell them apart. The only real difference is the no. of AF points (itself not really a major issue anyway)

The 7 remains the best AF I personally have used it's accuracy is excellent as is it's tracking ability.
The 5d is the weakest body I have AF wise..the 60's AF is unquestionably better than the 5d in both speed and low light.
At the time (2004) I would not have expected a body like the 60 to have a double cross sensor for AF it was not an expensive body.

In 2010 you could argue things have moved on enough to say mid level bodies should get one.

[twm47099]

I never used the 5. I was comparing the 60/70 with the 7 because it was the KM "Flagship" and sold in the US as a replacement for the 7 (wink, wink.)

I wasn't aware that the digitals were even "worse". I wonder what happened to the designers of the 7's AF? Even though the peripheral sensors on the 7 were not as good as the center double cross, they were better than either the "si" cameras or the cameras that came later.

tom

[bfitzgerald]

I have no idea what changed. I've spent the most time with the 5d and in comparison to previous film bodies it's rather "leisurely" regarding AF. It's just not turning the AF screw that fast. A200 added more speed and we heard about problems with Sigma lenses and the A700.

I'm not a very heavy AF user to be honest not into action a lot so the 5d's AF speed was not really a major problem. But something did change.

[Greg Beetham]

I'd say at a guess it might have been priorities with power supply, if I pick up the KM5D body in one hand and my Dynax60 in the other both with batteries inside, the KM5D is probably twice the weight of the Dynax60 even though the body size is similar, and why is this? I'd say electronic cameras need LOTS of power as compared to a film camera, I don't think digital cameras with all the features they have would even be a feasible endeavour without the advent of Lithium rechargable batteries.
So I think it was probably an acceptable balance at the time with so much power required for other things, the imaging sensor with it's fast response times, the processing unit, the large rear screen, Anit Shake and so forth, they decided to strike a medium performance level with AF to limit the amount of consumption....maybe.
Greg

[bfitzgerald]

I've done my own AF test just now and my results are as follows (no I didn't make a video but I might at some point)
50mm f1.7 fitted to
5d
Film 5
Dynax 60
A200
Film 7

Very low light small contrast target (A logitech logo on a speaker)
A200 could not lock
Km5d could not lock
Film 5 had a minor hunt then locked
Dynax 60 locked slightly quicker than the film 5
Film 7 obviously locked quickest (even with AF assist disabled)

I tried a slightly higher contrast subject and the A200 still did not lock, 5d has a moderate hunt then locked, 5 locked quickly, 60 quickly locked on and so did the 7

This is pretty low light now (as per my A200 v 5d high ISO shot) so obviously the A200 would be faster in good light. Low light wise the 5d and A200 are not great for AF without AF assist (of some type) All film bodies are better and by some margin too.

[R B Janes]

I'd always understood that the early digital models (7D, 5D and a100) has a lower-powered AF motor. Given that the 7D was about 5 times the price of a 7 on release, it is difficult to see why they cut corners on the AF motor.

The only reason I can reasonably come up with is that they were keen on making those early bodies as power-efficient as they could - if the lower-powered motor was also a little more power-efficient, it might justify its inclusion.

The upgrading of the AF motor coincided with the switch to the infolithium battery in the a200 and a700, which is what makes me suspect the power efficiency thing...

[bakubo]

I'd always understood that the early digital models (7D, 5D and a100) has a lower-powered AF motor. Given that the 7D was about 5 times the price of a 7 on release, it is difficult to see why they cut corners on the AF motor.

I bought a Minolta 7 in January 2002 for $599, but I think the price had been $699 a year or so earlier. I bought a KM 7D in March 2005 for $1499, but I think that was down from $1599 a bit earlier.

Yes, I agree though that it has been perplexing that AF took such a big step backward going from film bodies to digital bodies.

[David Kilpatrick]

There is a significant difference in the load>resistance>volt/amp supply, between the lithium cells used in the Dynax 7 (etc) and the lithium-ion rechargeables used the 7D, 5D etc. Power management, and the demands of the sensor and processor, probably dictated changes to the AF motor operation.

David