What is the best m43 for focus tracking?

Started Mar 31, 2013 | Questions thread
peevee1 Veteran Member • Posts: 6,247
Re: what is the best m43 for focus tracking?

clengman wrote:

peevee1 wrote:

clengman wrote:

JamieTux wrote:

Hey everyone, since selling my Canon gear to move to m43 I've been shooting with just one top quality body (I have a gf2 as well buits really lacking in dr for my style of shooting so one useful camera only scares me!) I had the money put aside to get the new improved AF all singing all dancing Oly when its announced but all seems quiet on that front at the moment.  So I'd like to get a complementary camera to go with my om-d.  If that is the top at the moment then I guess I will need to hire where needed until the new one is announced and make a decision then.

I've even been looking at the GH3 despite its size, but I can't find anyone willing to put their neck on the line and call a winner!

Thanks in advance for your thoughts!

I see the discussions of "focus tracking" a lot and I wonder when this is actually useful. As Mick, Anders and axlotl pointed out there's a big difference between C-AF (which seems to be very good in all the newer m43 cameras), and C-AF with tracking.

To take the commonly cited examples; sports, kids playing, and flying birds, why is the tracking part really useful? Why would you want to keep the camera absolutely still and rely on the camera to follow the subject. Seems to me that it's preferable to track the subject by panning the camera. As long as you can keep the AF box on your subject,

And this is a real problem often.

an E-M5 (or even an E-PL3 or G3 and all subsequent cameras from what I understand) should do just fine in C-AF mode at keeping whatever is in the focus box in focus.

C-AF with tracking seems like a separate issue. Whether it's a CDAF or PDAF camera it still requires some sophisticated software dedicated to object recognition. (i.e. determining that a particular patch of light projected on the sensor represents a discrete, permanent object and then following it even as it's color, luminance, shape and size change.) This is more a machine vision problem and not a focusing problem. Are there really any cameras that are good at that?

True, and PDAF is not necessary for object recognition. Just like you can recognize objects with one eye closed (without depth information), a proper algorithm can do it too. 3rd dimension can help a little in difficult conditions.

So you think it's kind of a depth map added to the 2D projection that helps with object recognition and tracking? I was just trying to think about it conceptually to get an idea why an SLR would have an advantage for tracking over a mirrorless camera. I can understand why cameras with PDAF would have an advantage for C-AF (though it seems like CDAF systems are definitely closing the gap.)

Yes. But I don't think it is even the biggest advantage decent Canon and Nikon DSLRs (and not ALL DSLRs) have in tracking. I heard a lot of info that other brands are not as good. Having low res low noise high frequency input from color array is one hardware help in addition to PDAF array. But I suspect the most lies in the long developed tracking algorithms, and that is where most advantage lies.

But if you develop a sensor with full-area binning, it is going to have better, lower noise signal than the color array. Hundreds of on-sensor PDAF pixels give better, more defined 3d info compared to few PDAF sensors in DSLRs. And algorithm development should be continued, I hope Oly and Pana have enough money to keep mathematicians and programmers working on tracking employed, despite the losses they suffer. But obvious delay in introducing of on-sensor PDAF compared to Nikon, Canon, Sony and Fuji is troubling (of course they need it least of all, and still better without it than anything but Nikon 1, but still).

But what is needed is massive pixel binning, turning output of 16 mpix sensor to something like 0.01 mpix first (similar to what color array on focusing sensors in DSLRs have). No camera processor can do object recognition on 24 Mbytes of noise raw data in real time yet even if a sensor could output that (and it cannot or even close, so line skipping is used, leading to low-res noisy picture). You would need a supercomputer for that.

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow