I just read the white paper about Canon's new on chip, phase detect, auto focus technology for the Canon 70D. It basically uses over 40 million pixel elements working under 20 million micro lenses to capture images and to do phase detection AF. Since the camera is dividing all the 20 million pixels in two we can expect two outcomes: 1. The accuracy of the phase detection AF on chip will be really good and the amount of light, cumulatively hitting the AF sensors, will be higher than that in more limited area systems (current Sony and m4:3rds). 2. The complexity of the processing and control of the sensor will be increased requiring more horsepower from the processors. I also suspect that having two photo diodes at each pixel site will have other, unexpected processing consequences. I'm sure Canon will work it out but it seems obvious that this is Canon's initial step (we, and the world at large, are, for the most part, ignoring their first failed effort...) in creating an answer to the mirrorless incursion. A new philosophy hoping to deal with the Mongol hordes of mirrorless offerings thundering across the plains of the consumer camera market.
I see this camera as a "proof of performance" sample. The next generation will move to a mirrorless configuration with an EVF because, if the on-chip AF is successful, there's no logical or economic reason to retain the more costly OVF. (To all those who say that cameras with EVFs are MORE costly I would say that you misunderstand. The new cameras may be more costly to YOU but they are much less expensive for the manufacturers and hence shore up receding profits. Not everything gets passed along to the consumer, especially not in a stale market...).
Presuming that Canon's 70D really performs (and the on chip AF is touted as being fully usable with every current Canon lens...) this means that both Canon and Nikon (in their J and V1 cameras) have proven to themselves that they need not have a secondary phase detect sensor integrated with finder optics and can offer a less expensive product at a steady price point to consumers who are acculturated and acclimated to doing most of their viewing and reviewing on big, rear screens. It also means, when Canon pull the curtain open on their mirrorless EVF iterations, that video gets better for most consumers because focus gets better for the video portion of the camera's feature set. And that's been a big source of unhappiness with Canon and Nikon amateur (and pro) video users who've come from faster focusing still systems.
The next step for all the makers is to finish coming to grips with fully electronic shutters. Once that's done we'll have taken out all the moving parts except for the control interfaces and that means faster cycling shot-to-shot and no wear and tear. Just in time to try and catch up with the mirrorless market that's already crowded in under the big tent.
Count on it.
I see this camera as a "proof of performance" sample. The next generation will move to a mirrorless configuration with an EVF because, if the on-chip AF is successful, there's no logical or economic reason to retain the more costly OVF. (To all those who say that cameras with EVFs are MORE costly I would say that you misunderstand. The new cameras may be more costly to YOU but they are much less expensive for the manufacturers and hence shore up receding profits. Not everything gets passed along to the consumer, especially not in a stale market...).
Presuming that Canon's 70D really performs (and the on chip AF is touted as being fully usable with every current Canon lens...) this means that both Canon and Nikon (in their J and V1 cameras) have proven to themselves that they need not have a secondary phase detect sensor integrated with finder optics and can offer a less expensive product at a steady price point to consumers who are acculturated and acclimated to doing most of their viewing and reviewing on big, rear screens. It also means, when Canon pull the curtain open on their mirrorless EVF iterations, that video gets better for most consumers because focus gets better for the video portion of the camera's feature set. And that's been a big source of unhappiness with Canon and Nikon amateur (and pro) video users who've come from faster focusing still systems.
The next step for all the makers is to finish coming to grips with fully electronic shutters. Once that's done we'll have taken out all the moving parts except for the control interfaces and that means faster cycling shot-to-shot and no wear and tear. Just in time to try and catch up with the mirrorless market that's already crowded in under the big tent.
Count on it.
In that first paragraph you've not quite understood what Canon is doing with the two photodiodes per pixel, Kirk.
ReplyDeleteThe two photodiodes per pixel are used for PDAF are used seperately when in focusing mode (as both are under a single microlens they look at different sides of the field ... just like a split prism finder) but both photodiodes are used for image capture (otherwise you'd throw away a stop of sensitivity and they didn't want to do that). That's the clever bit: all the light is used to capture the image.
The rest of you prognostication about EVFs and simplified cameras is right on the money (for economic reasons if no other).
I'm not sure that you are correct, Kevin. As I understand it: (from DPReview: "This offers 20.2MP resolution, but uses two photodiodes for every single pixel ('facing' left and right), so that they're all capable of on-chip phase detection." There are 20.2 million sits but each site or pixel has two photodiodes. That means that the pixels are all capable of on-chip phase detection but only one of the two photodiodes at each site is responsible for imaging. If this were not the case then the camera would be a 40.4 million pixel camera. Two photodiodes at each site. One for focus and one for imaging. Both under the same microlens. You can't have a photodiode continuously reading phase information and still sending simultaneous color information as well. But I could be misinterpreting. And, of course we need to add in the metering... Yikes. A lot of stuff going on on that chip. Maybe the photodiodes switch when the shutter is pressed but hmmmmm....
DeleteI'm correct, Kirk.
Delete"Maybe the photodiodes switch when the shutter is pressed".
Exactly! There is a AF mode they read two seperate photodiodes per pixel and a imaging mode they read both together (either they sum the charge or they read both and add them). It can even do this whilst taking video (though I suspect in that case small parts of the sensor are read (not the whole imaging part).
It can't ever be a 40Mpx camera because each pair of photodiodes is under one microlens One offset to the right "looks" left the other "looks right". That how the PDAF works you need to look in two directions and the difference between those directions (over a distance) tells you how far off the focus is. It's exactly the same a split prism. You actually need an array of pixels to do the AF (something Canon glosses over in their PR).
To image you use both of them together.
Watch the Canon video on vimeo and you'll see what's happening
http://vimeo.com/69519641
or read more here (a regular read for image sensor geeks)
http://image-sensors-world.blogspot.com/2013/07/canon-announces-dual-pixel-cmos-af.html
Cool. You and Gregg have convinced me of the error of my ways. I'll correct the article in the morning but the overall implications of this advance are obvious to me...
DeleteHi Kirk
ReplyDeleteI'm sure you're right that this tech will quickly find its way into a mirrorless body, let's hope that Canon's learnt some lessons from its last bomb in this sector, and in some respects I'm kinda glad I hung on to all those expensive lenses, but equally Canon remains at a big disadvantage with those very lenses, even if they do go the EVF body route - too big, too heavy and the IQ pay-off rarely seems worth it.
To help with your corrections, here is the DPR explanation:
ReplyDeleteCanon's 'Dual Pixel CMOS AF' sensor looks at the concept of on-chip phase detection, and takes it to a logical conclusion. Instead of masking some pixels so they only 'face' left or right, it uses two photodiodes for each pixel, so every single one has a left- and right-facing component that can be used for phase detection. When a photograph is taken, the output from the two photodiodes is combined.
So we should not confuse pixels with photo diode. With most sensors it's a 1-to-1 relationship, but with the new Canon it's 2 photo diodes for each pixel. One photo diode "looks" to the left side of the pixel, the other right and voila, phase detection AF. The luminescence value and Bayer color are combined and the sum is used for picture taking.
I best we see this in the EOS-m mark 2.
Cheers
John Driggers
Adelaide South Australia
I am betting the 7DmkII gets the EVF.
ReplyDeleteI see this as a big deal in terms of getting the mirror out of dSLRs, and of less importance for video though it will certainly help there too.
ReplyDeleteI question how many of the target photographers of the 70D will produce better videos because of the AF, compared to other mirrorless systems which can already af during video AND allow shooting in an EVF. That said, accurate follow focus could really bring some polish to amateur video without the learning curve...
Bring on the EVFs! Having used EVFs alongside OVFs for 4 years, I know feel like I am stepping backwards when going from EVF to OVF. I get so much more information via the EVF... About the only time I prefer an OVF is when shooting sports. But that will go away (almost surely) in time as processing power and faster refresh rates improve.
ReplyDeleteThis technology, at the very least, has bought Canon more time in the fight against mirrorless.
ReplyDeleteOne word: yup. More words: if this sensor design lives up to its promise, Canon's mirror-less cameras, then Canon is going to absolutely smoke Nikon's own mirror-less efforts (and do it with added extra L glass).
ReplyDelete
ReplyDeleteThanks for review, it was excellent and very informative.