[Duplo]

There's always room for improving the AF performance, but I think there's a lesson to be learned in giving the end user TOO much stuff to customize.

Once you have gotten your head around it, it is not too bad, but there is a lot of work setting it up initially and figuring it out, a number of memory banks actually make for a fairly simple switch between different groups of settings, making it almost as simple to use, albeit tailored to your needs.

But I hear you in the way that there is a breaking point and I think the D3(x)/1DsIII are getting close to it, without surpassing it though

see, that's what i hate about pentax users. no matter how inflamed the subject, they always know how to spoil a good fight and somehow get along. so reasonable and rational. damn.

Well we are a civil bunch in here

seriously, i believe there is of course room for improvement, but i think the lack of speed is overrated, i think even the screwdrive system would be able to perform a lot better if it was smarter at tracking, that's what most people complain about if i understand correctly. it's not even that hard, especially with 9 proper sensors already in there, i believe a bit of firmware intelligence might improve things enough to make many people happy (it will not turn the k10d/k20d into a d300, but that's not the point), i also dare to believe a chunkier and dedicated cpu for af would ofcourse help, but is not required (i have a hunch nikon/canondo it "wrong", meaning they can afford the resources and hardware, so they just throw it in there, it can be done smarter with really simple algorythms and some thought). and i come back to the idea of (partially) opensource firmware, i guarantee that we would be scratching our heads in about a year "how the hell was that possible, we all thought the safox 8 was doomed". just try it, pentax..

Now here I think you are wrong thugh, not in the sense that the Safox VIII cannot be improved a lot by enhanced programming, I think we agree on that.

But I disagree with you, when you say that a dedicated AF sensor is the wrong direction.
I am pretty sure that the reason you see dedicated processors for AF alone in the single digit canon/nikon bodies is because it is the fastest way to get all the way up, the D300 while not a slow AF body, is not as fast as the D3/D3, so even if you have excellent programming, then giving it its own processor means a speed increase and that it will perform unaffected of how much processing the rest of the camera needs, thus providing a much more consistent AF performance.

Thus a dedicated AF processor is a way to ensure stability and consistency, not to mention additional speed as it will operate independant of the rest of the cameras processing requirements.

[res3567]

Here is my post on the subject over on the darkside forum:

Canon Digital Photography Forums - View Single Post - Should I switch to Pentax?

Wow.

Great comparison.

I too own Canon equipment and a member of that Fornm.

I have two P&S's ( my G 3 is my backup to my K10D And have an A75)

I also have an EOS A2 with a Sigma 28-105 f2.8-4.0.

No DSLR.

Shot my cousin's wedding with hs 30D.

Compared my K10D to his 30D.

I came up with mostly the same conclusions as you did especially when it came to AF.

I rarely visit the Canon forum.

I live hear and have really stacked my equipment based on this forum.

Thank you all for your help!

[nanok]

thomas, if you read what i said carefully, you will see that i do not disagree that a dedicated af unit is a [edit] good choice, just that it might be possible to do it smarter, and move to the dedicated af module after. i have a hunch the way the better af (with the dedicated unit) is achieved on the big bucks bodyes is wastefull in a way (i might, ofcourse, be wrong), as i said above, doing it only "in software" will not instantly turn the k10d/k20d into speed daemons, but it will help a lot, more hardware (and firmware intelligence) would probably be required to get close to state of the art performance. actually, i believe a dedicated af module of a certain sort would be the best way to go, but of a certain sort.. (nobody will ever do what i have in mind though, not in this day and age).

anyway, to put it briefly, we do agree on most points

[OregonJim]

But since my 401K is worthless now and I no longer need to save for retirement, since I will never be able to retire...
I can now afford to switch systems if the new Pentax is not a Paradigm Shift in AF reality...

Yeah, my 401K went down the tubes, too. I like your twisted yet somehow logical form of rationalization...



EDIT: Arrghh - just realized that I'm contributing to keeping this thread alive...

[res3567]

Yeah, my 401K went down the tubes, too. I like your twisted yet somehow logical form of rationalization...



EDIT: Arrghh - just realized that I'm contributing to keeping this thread alive...

Yea Jim. It just keeps going and going and going......

It is alive and waiting for fresh meat daily.

Use old timers on this thread have to look to see who got pulled in and we end up respondeing when we feel we have two.

[*isteve]

Thus a dedicated AF processor is a way to ensure stability and consistency, not to mention additional speed as it will operate independant of the rest of the cameras processing requirements.

I think for continuous AF, this is the REAL key issue.

The camera MUST continue to reassess focus and turn the lens even when the CPU is busy processing the sensor data. In computer speak, this is offloading, like a separate graphics card with its own memory. Its frees up CPU cycles on the main processor, and frees up bandwidth on the main bus. This would make the whole camera faster, not just the AF.

I suspect all the 52 point Nikon cameras have a dedicated processor, but I suspect there are two versions, one relatively slow with cache only, and one really fast with dedicated RAM or a larger cache.

Another issue for the D300 and D700 is the mirror return speed and blackout time. The lower the blackout time, the more time the AF system is tracking. The D3 if I remember has a VERY short blackout time, something that cannot be said of a K20D.

[Duplo]

I think for continuous AF, this is the REAL key issue.

The camera MUST continue to reassess focus and turn the lens even when the CPU is busy processing the sensor data. In computer speak, this is offloading, like a separate graphics card with its own memory. Its frees up CPU cycles on the main processor, and frees up bandwidth on the main bus. This would make the whole camera faster, not just the AF.

I suspect all the 52 point Nikon cameras have a dedicated processor, but I suspect there are two versions, one relatively slow with cache only, and one really fast with dedicated RAM or a larger cache.

Another issue for the D300 and D700 is the mirror return speed and blackout time. The lower the blackout time, the more time the AF system is tracking. The D3 if I remember has a VERY short blackout time, something that cannot be said of a K20D.

I am somehow not surprised that you and I agree on this stuff and what you say above sounds entirely resonable.

To sum it all up it seems to come down to 4 areas.

Programming
dedicated processing circuits
Shutter lag
mirror black out

and oddly enough I find the latter two much more important to me

[nanok]

I think for continuous AF, this is the REAL key issue.

The camera MUST continue to reassess focus and turn the lens even when the CPU is busy processing the sensor data. In computer speak, this is offloading, like a separate graphics card with its own memory. Its frees up CPU cycles on the main processor, and frees up bandwidth on the main bus. This would make the whole camera faster, not just the AF.

I suspect all the 52 point Nikon cameras have a dedicated processor, but I suspect there are two versions, one relatively slow with cache only, and one really fast with dedicated RAM or a larger cache.

Another issue for the D300 and D700 is the mirror return speed and blackout time. The lower the blackout time, the more time the AF system is tracking. The D3 if I remember has a VERY short blackout time, something that cannot be said of a K20D.

those are all very good points, but, at the risk of being a pita, i do not care so much about processing overhead, offloading and such when processing the image data, let alone af sensor blackout (of any kind) -- which, btw, can be and probably is handled in all cameras by guessing -- and for the same reason a big chunky cpu to handle that does not concern me either: as far as i could tell, the pentax cameras would have no use for it, and pentax was right not to include one, the problem is not that the k10d/k20d loses track of the subject during mirror blackout, or that it lags behind while processing the image data in burst mode, the problem is that it does no actual tracking, none, whatsoever, it just.. focuses continuously (please correct me if i am wrong here, i don't use af-c so much, so perhaps i am missing something essential). so while you are ofcourse right, it is in this case like talking about tuning the psi of the turbo on the engine, before actually having a turbine to speak of, i am under the impression that some careful thought and development on _how_ to do the tracking would be the first and greatest step pentax could do to make everybody happy, and the results would be very satisfactory for an incremental step, if this step is done right, the best way possible, there is the potential to leapfrog both canon and nikon in a near future, by adding the right hardware and developing the algorythms further. all i am saying is: one step at a time

[KungPOW]

Sorry If i missed it in this thread,

Does Pentax have predictive autofocus?

[palmor]

Sorry If i missed it in this thread,

Does Pentax have predictive autofocus?

Nope



John

[nanok]

while this is all true, and the pentax af system is obviously a device from hell/the stone age/insert-favourite-insult-here, i just realized that the most interesting shots i took this week are of one of the sparkplugs that i just removed from my car. now, i know something is wrong with this picture, but i can't put my finger on it (and no, it's not the safox viii holding the creative juices from flowing).

anyone care to see my old ngk spark plugs? :/

[*isteve]

those are all very good points, but, at the risk of being a pita, i do not care so much about processing overhead, offloading and such when processing the image data, let alone af sensor blackout (of any kind) -- which, btw, can be and probably is handled in all cameras by guessing -- and for the same reason a big chunky cpu to handle that does not concern me either: as far as i could tell, the pentax cameras would have no use for it, and pentax was right not to include one, the problem is not that the k10d/k20d loses track of the subject during mirror blackout, or that it lags behind while processing the image data in burst mode, the problem is that it does no actual tracking, none, whatsoever, it just.. focuses continuously (please correct me if i am wrong here, i don't use af-c so much, so perhaps i am missing something essential). so while you are ofcourse right, it is in this case like talking about tuning the psi of the turbo on the engine, before actually having a turbine to speak of, i am under the impression that some careful thought and development on _how_ to do the tracking would be the first and greatest step pentax could do to make everybody happy, and the results would be very satisfactory for an incremental step, if this step is done right, the best way possible, there is the potential to leapfrog both canon and nikon in a near future, by adding the right hardware and developing the algorythms further. all i am saying is: one step at a time

The issue is that when the camera stops tracking because its busy it has to "reacquire" the target when its no longer busy. This takes even more time and increases the chances of losing it altogether.

For a small subject at close distances even a moderate rate of motion can change thd focus point quite rapidly. In AFC mode that means a real possibility the target will not have been acquired when pressing the shutter.

[nanok]

steve, as i said (i did mean it), i agree, yours are very good points. however, i must admit i have too little experience with af-c (i'm just too old fashioned i guess), so my first question is: does the k10d or k20d appear to be doing any "guess work" while in af-c, so called predictive af, or is it just doing repeated af-s, at very high frequency? it was my impression that it does the later , or that, if it does some guessing, well... it's not very smart guessing. so my impression is that it is using brute force for af-c, instead of trying to do it the smart way, and if that is true, the improvement with the current "force" available with the current af system if trying to do it smarter (more economical) would be rather big. that was my point.

it is not arguable if the blackout, processor overhead (when using the same processor as for image processing/i/o to the storage) are very important for high-end af-c tracking, but are we even at that point?

i don't mean to bash the pentax af (i think it's had enough), i am jsut trying to constructively figure out what might be done to improve it, for what it's worth (to be honest, as far as i am concerned, the pentax af does great: it does a hell of a better job than i can, and faster, when i let it do it -- not very often -- so i guess i can't complain, i am one of those guys who got af when switching to digital as part of the deal, and said "oh well, i'm depending on batteryes now anyway, so what the hell.." )

[pingflood]

those are all very good points, but, at the risk of being a pita, i do not care so much about processing overhead, offloading and such when processing the image data, let alone af sensor blackout (of any kind) -- which, btw, can be and probably is handled in all cameras by guessing -- and for the same reason a big chunky cpu to handle that does not concern me either: as far as i could tell, the pentax cameras would have no use for it, and pentax was right not to include one, the problem is not that the k10d/k20d loses track of the subject during mirror blackout, or that it lags behind while processing the image data in burst mode, the problem is that it does no actual tracking, none, whatsoever, it just.. focuses continuously (please correct me if i am wrong here, i don't use af-c so much, so perhaps i am missing something essential). so while you are ofcourse right, it is in this case like talking about tuning the psi of the turbo on the engine, before actually having a turbine to speak of, i am under the impression that some careful thought and development on _how_ to do the tracking would be the first and greatest step pentax could do to make everybody happy, and the results would be very satisfactory for an incremental step, if this step is done right, the best way possible, there is the potential to leapfrog both canon and nikon in a near future, by adding the right hardware and developing the algorythms further. all i am saying is: one step at a time

You're making a rather curious assumption here: that both Canon and Nikon are just relying on expensive hardware and inefficient AF approaches. Believe me, both companies have engineers scrambling to figure out how to make the systems smarter so that they can stay ahead. Nobody is going to have an "aha" moment at Pentax and leapfrog everyone else with their sheer brilliance while Canon and Nikon crudely throw more CPU power at the problem. Careful thought and development? You bet that's been done to death.

[ManuH]

To sum it all up it seems to come down to 4 areas.

Programming
dedicated processing circuits
Shutter lag
mirror black out

And lens AF speed (SDM or screw). BTW speaking of the lenses the in-lens stabilsation also favors AF speed: the image is stabilised so there is less work for the AF sensors.

Pentax AF-C can also be made slightly faster if one is using the central point only. When using the auto-points it's slower. That may confirm the weak non-dedicated to AF CPU theory.