[falconeye]

so we're talking about a distant future; OK then, I'm more concerned with the near future, and I don't see the contrast detection AF

I see your point and it may be the reason why phase still beats contrast. However, a full density 60fps read-out could already be done for a center focus region rectangle today (and faster even in the future). It is just not what vendors decided to be their priority.

The EVIL cameras may change this as soon as there are dedicated sensors for EVIL (e.g., with lag-free full resolution zoom capability). So, future may be a not so distant one.



BTW, Samsung seems to call EVIL cameras "SLD" thereby making reference to "SLR". SLD shall mean "single lens direct".

[juu]

SLD is actually a good acronym. I like it.

[Kunzite]

falconeye, we shall see CD-AF will improve, that's for sure (and we all know Pentax cameras are nowhere near the best CD-AF implementation)
One question: do you think dedicated (e.g. made with CD-AF in mind) AF lenses are required for fast CD-AF? What about our SDM lenses? Or even screw driven ones?

SLD looks like "SLR wannabe" for me

[falconeye]

One question: do you think dedicated (e.g. made with CD-AF in mind) AF lenses are required for fast CD-AF?

No, I don't think so.

However, the ability to precisely shift focus by small amounts will become very important, i.e., lenses with no empty play between forward and backward movements. That's more a question of fine mechanical precision than anything else.

Not because CD-AF needs it but because the increased focus accuracy made possible by CD-AF would require it. At f/1.4 and with 5 µm pixels, the required positioning accuracy is 0.007 mm. Which is only 0.12° or 7arcmin or 1/14mm turning distance (the width of a hair -- on the focus scale!) of the focus ring for the FA 31 Ltd. The FA 31 Ltd may have this accuracy anyway.

Of course, CD-AF won't be faster than the lens can shift focus. But that's not different from now. Only bad AF algorithms do shift forth and back (as they do now (**)) and need special lenses to hide this defect. This requirement will fade away. What remains are micro steps forth and back at the beginning and end of AF which you won't even notice with an accurate enough lens and enough ambient light (*).

__
(*) A lack of ambient light means that the contrast function is noisier and larger focus steps are required to obtain significant delta measurements which then may become noticeable. But CD-AF should be able to focus where it is too dark for phase AF.

(**) Current CD-AF hunts forth and back in a way I would never let my engineers pass thru. I know it can be done better. It is embarrassing. The Samsung NX is somewhat better but not enough.

---

Last edited by falconeye; 02-04-2010 at 07:51 AM.

[johnmflores]

Falconeye, are you offering college credits for this thread? Thanks for sharing your vast knowledge - it makes it worth slogging through the pie throwing contest that's occurring simultaneously.

[reflexio]

Is it likely that my 14mm 2.8 prime would work with FF sensor?

I know that it effectively is 21mm at the moment, but if I got to a full 14mm with a FF sensor, that'd be cool.

[Wheatfield]

For the record, I am a working professional photographer, educator, and trainer. I've seen many such marketing tactics over the years influence the general public and consumer base at large. I have learned not to make any assumptions as to the validity of a metric without knowing the source.

We seem to have similar backgrounds.
I've been a professional mentor for a very long time now.
I too am leery of marketing, but I will look at a new or improved technology without a jaundiced eye just because the ad guys like it.
As for not clamoring for ISO 6400 back in the day, the introduction of a usable 400 ISO colour film was cause for great joy, and there was always interest in getting faster colour films, or finer grained fast B&W films.
There has always been a contingent of photographers who push process film to try to get more speed out of it.
Without the internet, we clamored more quietly, I guess.
I suspect if the internet had existed in the 1960s, we'd be bemoaning the failings of Kodachrome and wishing for an ISO 64 version.
Some guy would be ranting about the new E-4 films and how they seemed to fade if left in the sun.
But I digress.
And we'd be arguing about the Zone System, which in a sense is what we all practice now if we do histogram sensitive post processing.

[Miserere]

Is it likely that my 14mm 2.8 prime would work with FF sensor?

Is it the DA 14mm f/2.8? In that case, no, it won't work on FF. For that FoV on FF you could get the FA 20mm f/2.8.

.

[viggen]

Is it the DA 14mm f/2.8? In that case, no, it won't work on FF. For that FoV on FF you could get the FA 20mm f/2.8.

.

Or it could work if the sensor auto-crops to get an equivalent APS-C captor size.

[Miserere]

Or it could work if the sensor auto-crops to get an equivalent APS-C captor size.

Well yes, but what I meant was that it wouldn't cover the whole area of the FF sensor. If Pentax release a FF they'll have to allow for use of DA lenses on the center portion of the sensor.

.

[*isteve]

The major advantage of mirrorless cameras as you suggested is the lack of any need for extensive callibration because the focus system uses direct feedback from the sensor. Effectively, these systems can be made as accurate as the tolerances in the lens (slack in the AF mechanism) can allow for.

Speed is an interesting issue. For me accuracy is far more critical. The problem with the "double take" as far as I see it is that you have to keep the same object in the focus point fairly precisely, so you have to make it fast and you have to keep the camera still, which will require lens engineering on a new level.

Backward compatibility may therefore be an issue, as with the micro 4/3 mount,

No, I don't think so.

However, the ability to precisely shift focus by small amounts will become very important, i.e., lenses with no empty play between forward and backward movements. That's more a question of fine mechanical precision than anything else.

Not because CD-AF needs it but because the increased focus accuracy made possible by CD-AF would require it. At f/1.4 and with 5 µm pixels, the required positioning accuracy is 0.007 mm. Which is only 0.12° or 7arcmin or 1/14mm turning distance (the width of a hair -- on the focus scale!) of the focus ring for the FA 31 Ltd. The FA 31 Ltd may have this accuracy anyway.

Of course, CD-AF won't be faster than the lens can shift focus. But that's not different from now. Only bad AF algorithms do shift forth and back (as they do now (**)) and need special lenses to hide this defect. This requirement will fade away. What remains are micro steps forth and back at the beginning and end of AF which you won't even notice with an accurate enough lens and enough ambient light (*).

__
(*) A lack of ambient light means that the contrast function is noisier and larger focus steps are required to obtain significant delta measurements which then may become noticeable. But CD-AF should be able to focus where it is too dark for phase AF.

(**) Current CD-AF hunts forth and back in a way I would never let my engineers pass thru. I know it can be done better. It is embarrassing. The Samsung NX is somewhat better but not enough.

[viggen]

Well yes, but what I meant was that it wouldn't cover the whole area of the FF sensor. If Pentax release a FF they'll have to allow for use of DA lenses on the center portion of the sensor.

.

Yes, of course; I just meant that DA lenses won't suddenly stop working with an hypothetic Pentax FF body (which is what one could be lead to believe).

As stupid as my remark is (I'm quite conscious of it, don't worry), it is something worth saying : a FF body doesn't mean that suddenly, all DA lenses are worthless.
Sure, it's better to have FF lenses to go with a FF body.

Or could there be advantages using a DA (or assimilated) lens considering the iso behavior (as the pixel count would be lower), or the dynamic range?


Sorry for being slightly off-topic.

[Class A]

Phase detect is a special form of contrast detect. One where the aperture has two/four holes f/5.6 apart and where contrast is only computed over a small region of the image.

Falk, do you have pointer to some material about the details of phase detect AF?

I was under the assumption that phase detect AF works by observing an offset between patterns (different projections of the same small portion of an image, -> split prism principle), not by detecting contrast within a pattern.

Of course, you could try to both reduce the offset to zero and maximise the contrast at the same time, but doesn't the term "phase" in "phase detect" imply that the main principle is to observe a phase shift, i.e., offset, and then adjust focus until the offset is below a certain threshold?

[falconeye]

I was under the assumption that phase detect AF works by observing an offset between patterns, not by detecting contrast within a pattern.

I am sorry that I didn't make the distinction. In my rather abstract vocabulary, both are the same.

Phase shift maximizes the correlation function c(dx) between two (almost one dimensional) images a, b, shifted by distance dx.

Contrast maximizes the auto correlation function c(dx) of a (two dimensional) image a with itself, shifted by a focussing distance dx.

That's the same in my book, i.e., some correlation function is maximized.

The good point about such an abstract point of view is that it becomes possible to reason about when one variant beats the other, in a theoretic way.


Disclaimer: A good contrast detect AF system would use a function different from the auto correlation function. For my own algorithm I studied a variety of options. But this wouldn't matter for the abstract discussion I am having here.

[Class A]

That's the same in my book, i.e., some correlation function is maximized.

I fully agree with this (very useful) abstract view.
I was puzzled by the wording "contrast calculation" in the context of phase detection AF, but I now see how this wording sort of makes sense using the abstract view.