Canon Confirms Development of High Megapixel Camera

[Lee Jay]

I have read this theory before as well. I would love to see them come up with this sort of thing. Flip a switch and turn this setting when the need arises.

Maybe that 150,000 pixel metering system could decide automatically if the scene contrast was high enough to need it.

 

[PureClassA]

I have read this theory before as well. I would love to see them come up with this sort of thing. Flip a switch and turn this setting when the need arises.

Maybe that 150,000 pixel metering system could decide automatically if the scene contrast was high enough to need it.

I have on occasion done HDR shots on my 6D and 16-35 f4 L. I've done both "in camera" with a resulting JPEG and by simply bracketing 3 RAW exposures usually around +/- 2 stops and then HDR combining in Photoshop. More often than not I am happier with the results I get from pushing shadows in a single shot from the group. I've never tried the DUAL ISO ML hack. How does that work? Is it a simultaneous exposure?? Or is it a bracketed composite more like traditional HDR?

 

[Lee Jay]

I have read this theory before as well. I would love to see them come up with this sort of thing. Flip a switch and turn this setting when the need arises.

Maybe that 150,000 pixel metering system could decide automatically if the scene contrast was high enough to need it.

I have on occasion done HDR shots on my 6D and 16-35 f4 L. I've done both "in camera" with a resulting JPEG and by simply bracketing 3 RAW exposures usually around +/- 2 stops and then HDR combining in Photoshop. More often than not I am happier with the results I get from pushing shadows in a single shot from the group.

My experience as well.

I've never tried the DUAL ISO ML hack. How does that work? Is it a simultaneous exposure?? Or is it a bracketed composite more like traditional HDR?

It's one exposure, but half the pixels are read at a high ISO and half are read at a low ISO.

http://www.magiclantern.fm/forum/?topic=7139.0

 

[PureClassA]

Oh well then the question has been answered. Obviously firmware can be built to do it. But breaking up a 22mp sensor vs 50mp sensor obviously demands far more CPU power. But assuming the Digic 6 (or perhaps 7 in a 5D4 or whatever this is) can. Very interesting. I'd love to find a RAW shot done with this ML HDR hack.

I'll have to read the article. I always had issues with multi shot HDR because "landscapes DO move" even the slightest breeze moves leaves and branches and blades of grass and aligning all that to recover a sharp multilayer shot is a beast

 

[PureClassA]

Ok now I'm even more perplex. The blog mentions "taking advantage of the full 14 stops of DR the sensor IS CAPABLE of" So now I gotta ask....if the sensor is CAPABLE of it....what the hell are we talking about with bad DR? Granted the base Canon firmware doesn't allow for dual ISO HDR single shooting... but it seems the sensor CAN produce it given the right tweak. I guess I'm wondering if the DR issue is really a sensor issue...or a Canon firmware issue? I need people way better versed on this than I am to help me. Glad ya'll are here LOL

 

[Lee Jay]

Ok now I'm even more perplex. The blog mentions "taking advantage of the full 14 stops of DR the sensor IS CAPABLE of" So now I gotta ask....if the sensor is CAPABLE of it....what the hell are we talking about with bad DR?

The sensor can produce it, but the way the data is read out of the sensor is noisy enough to raise the noise floor a couple of stops from that of the sensor itself.

 

[photonius]

Let me propose, again, another approach Canon could potentially use.

Let's say this camera has 7DII pixels, and 52 million of them.

That means it has 104 million separate pixels because of the dual pixel design of the pixels.

The pixels can obviously be read separately or they wouldn't work for phase detection focusing in live view and video.

So, let's say the camera has a mode where the two halves of each pixel are read at different ISOs. For sake of argument, let's say ISO 100 and ISO 1600 are used.

The result would be that the top four stops would have twice the shot noise they would have if you didn't do that because half of your data is clipped there. But who cares? The top four stops have so much signal that it really doesn't matter. The overlapping range would be the same. The bottom of the range would have something like 3-4 more stops of shadow performance due to the far lower read noise at ISO 1600.

Combine these two in a way similar to the way Magic Lantern dual-ISO works and you have a 15 stop or so DR image at 52MP.

Now, that 15 stop image won't fit into a 14 bit raw so you generate a 16 bit raw from your 14 bit * 2 raw data sets.

Now you have 80+MB raw files, give or take, with 52MP and 15 stops of base-ISO DR.

i proposed that already a long time ago when the 70d came out.
It seems canon is aware of this, but maybe cant make it work that easily (from some cryptic comments in interviews)
They may have even tried for the 7D II (causing the delay, extra firmware update for 7D).

 

[PureClassA]

Ahhh hence all the talk from jrista and others about Canon's "read noise" problem. Gotcha. And as far as the 70D goes.... if ML can do it on a non-dual pixel format, I would think could easily do it a DPAF system. It's the IDEAL sensor type to pull this very thing off. And a DPAF 52MP sensor...well good lord. Sounds like Canon already has the ability to smoke out serious DR from the sensor, but the noise from reading it is the issue. I'm catchin up. Thanks

 

[Loswr]

I always had issues with multi shot HDR because "landscapes DO move" even the slightest breeze moves leaves and branches and blades of grass and aligning all that to recover a sharp multilayer shot is a beast

No, landscapes don't move. It must be true, dgatwood said so. :

 

[jrista]

Ok now I'm even more perplex. The blog mentions "taking advantage of the full 14 stops of DR the sensor IS CAPABLE of" So now I gotta ask....if the sensor is CAPABLE of it....what the hell are we talking about with bad DR? Granted the base Canon firmware doesn't allow for dual ISO HDR single shooting... but it seems the sensor CAN produce it given the right tweak. I guess I'm wondering if the DR issue is really a sensor issue...or a Canon firmware issue? I need people way better versed on this than I am to help me. Glad ya'll are here LOL

Most SENSORs these days are not that noisy in the grand scheme of things. If Roger Clark's data is anything to go by, Canon sensor DR is anywhere from 14.3 to 15.7 stops. Canon's most significant problem is the introduction of noise from downstream (off-die) electronics. Primarily, read noise introduced by their high frequency ADC units, of which there are four, eight or sixteen separate units on the PCB board, between the sensor and the DIGIC processor(s). It's that electronics that adds some quantity of noise to every single image read off the sensor.


If the sensor starts out with 14-15 stops of DR, it's the introduction of read noise from other non-sensor electronics that is basically "eating away" at dynamic range. Canon's off-die electronics, given that their sensors usually have around 11 stops of DR, are apparently reducing dynamic range by 3-4 stops! Canon's problem isn't necessarily their sensors (not explicitly)...it's the way they read the sensor signal out. The solution to Canon's problem is to reduce the operating frequency of readout electronics, and reduce trace paths along which pixel data is transferrred. The best way to do that IS to improve their sensors...by moving ADC units onto the sensor die, by increasing ADC parallelism, and by employing other technologies allowed by CMOS sensor designs to increase Q.E. and reduce sources of additional noise (outside of the amplifiers in each pixel themselves.)


That's what Sony did...they moved all the readout electronics onto the sensor, hyperparallelized the ADC units (one per column), and employed a number of other techniques to reduce noise futher (i.e. moving high frequency components, such as the clock, to remote areas of the die so they don't risk injecting additional noise into the pixel data as it's read out, use of digital CDS, use of per-column tuning in each ADC unit to eliminate vertical banding, etc.)

 

[Loswr]

As private has stated, when shooting hundreds of images, even thousands maybe, massive files are a pain.

Then people need to get better and judging what will/won't be a good photograph before they press the shutter release button and also get better at deleting the deritus from their photograph catalogue.

That is a stupid thing to say.

I often work composite files with 50-100 RAW files in them, each and every one of them adds something none of the others do. It is not uncommon for me to shoot 200 images on a tripod for a key image.

So in this case your keeper rate is 1 in 200. Why then do you need to keep the other 199?

Maybe if you'd read a little more carefully, you'd have discovered your question had already been answered...

I often work composite files with 50-100 RAW files in them, each and every one of them adds something none of the others do. It is not uncommon for me to shoot 200 images on a tripod for a key image. I have to shoot many angles and keep all of them because i never know what my clients will want altered.

 

[Don Haines]

There's a lot of talk about this camera being a 5DIV. In Canon Rumors post from November, it's stated that the new high megapixel camera will come in a new line, above the 5D line, which makes sense to me. The 5D3 is pretty much the ideal all-rounder camera and I can't see Canon messing with what has been a winning formula.

http://www.canonrumors.com/2014/11/another-50mp-ff-dslr-mention-cr2/

Such a camera would be even more of an "all-rounder" giving the reach of a 7DII, the landscape potential of medium format

You're wrong about that. The DR and noise characteristics of a scaled up 7DII sensor would not put it on a par with other landscape cameras.

I'm still waiting to see my first real-world landscape image where the amount of DR you can get from a Canon sensor at base ISO is insufficient. All I ever see is contrived scenes or scenes where compressing all that DR into the final image makes it look like crap.

Here's an easy one for you: go and shoot a sunset (facing the sun) where you've got interesting things in the foreground that are in shadow. And yes, compressing DR into a final image does

Sunset? Sunset? That's overkill.
Here is an owl that was clipped on both ends of the DR scale... you can run out of DR on a cloudy day....

 

[PureClassA]

Ok now I'm even more perplex. The blog mentions "taking advantage of the full 14 stops of DR the sensor IS CAPABLE of" So now I gotta ask....if the sensor is CAPABLE of it....what the hell are we talking about with bad DR? Granted the base Canon firmware doesn't allow for dual ISO HDR single shooting... but it seems the sensor CAN produce it given the right tweak. I guess I'm wondering if the DR issue is really a sensor issue...or a Canon firmware issue? I need people way better versed on this than I am to help me. Glad ya'll are here LOL

Most SENSORs these days are not that noisy in the grand scheme of things. If Roger Clark's data is anything to go by, Canon sensor DR is anywhere from 14.3 to 15.7 stops. Canon's most significant problem is the introduction of noise from downstream (off-die) electronics. Primarily, read noise introduced by their high frequency ADC units, of which there are four, eight or sixteen separate units on the PCB board, between the sensor and the DIGIC processor(s). It's that electronics that adds some quantity of noise to every single image read off the sensor.


If the sensor starts out with 14-15 stops of DR, it's the introduction of read noise from other non-sensor electronics that is basically "eating away" at dynamic range. Canon's off-die electronics, given that their sensors usually have around 11 stops of DR, are apparently reducing dynamic range by 3-4 stops! Canon's problem isn't necessarily their sensors (not explicitly)...it's the way they read the sensor signal out. The solution to Canon's problem is to reduce the operating frequency of readout electronics, and reduce trace paths along which pixel data is transferrred. The best way to do that IS to improve their sensors...by moving ADC units onto the sensor die, by increasing ADC parallelism, and by employing other technologies allowed by CMOS sensor designs to increase Q.E. and reduce sources of additional noise (outside of the amplifiers in each pixel themselves.)


That's what Sony did...they moved all the readout electronics onto the sensor, hyperparallelized the ADC units (one per column), and employed a number of other techniques to reduce noise futher (i.e. moving high frequency components, such as the clock, to remote areas of the die so they don't risk injecting additional noise into the pixel data as it's read out, use of digital CDS, use of per-column tuning in each ADC unit to eliminate vertical banding, etc.)

So then my next question would be "Can Canon not employ a similar design? Is such a design patented by Sony?" "If they can, what's stopping them from doing this if a solution has already be discovered? Would putting the ADC right on the chip be some radically expensive redesign?"

 

[Maiaibing]

Such a silly comment is hardly worth responding to. Canon understands their market extremely well. There is ample evidence of that for anyone paying the least bit of attention.

Since a Canon Executive has just given an interview saying that they missed their customer's need for a high megapix camera its fair to say Canon think's they actually got the market sentiment wrong on this count.

Also, Canon DSLR sales are not only plummeting but also doing so at a much faster rate than Canon expected only 12 months ago (leading to two downwards revisions of their estimated 2014 DSLR sales during this year).

Doubt the Canon board members are as impressed as you are.

 

[Lee Jay]

There's a lot of talk about this camera being a 5DIV. In Canon Rumors post from November, it's stated that the new high megapixel camera will come in a new line, above the 5D line, which makes sense to me. The 5D3 is pretty much the ideal all-rounder camera and I can't see Canon messing with what has been a winning formula.

http://www.canonrumors.com/2014/11/another-50mp-ff-dslr-mention-cr2/

Such a camera would be even more of an "all-rounder" giving the reach of a 7DII, the landscape potential of medium format

You're wrong about that. The DR and noise characteristics of a scaled up 7DII sensor would not put it on a par with other landscape cameras.

I'm still waiting to see my first real-world landscape image where the amount of DR you can get from a Canon sensor at base ISO is insufficient. All I ever see is contrived scenes or scenes where compressing all that DR into the final image makes it look like crap.

Here's an easy one for you: go and shoot a sunset (facing the sun) where you've got interesting things in the foreground that are in shadow. And yes, compressing DR into a final image does

I did that once, and I didn't have enough DR. However, a little math showed that I needed 30 stops of DR for that shot. So 1 or 2 extra stops from a Sony sensor would have done me exactly no good. Even bracketing didn't get the job done.

I've never yet found a shot that needed just a little bit more base ISO DR than I can get from a Canon sensor, but not too much for a Sony sensor.

 

[jrista]

So then my next question would be "Can Canon not employ a similar design? Is such a design patented by Sony?" "If they can, what's stopping them from doing this if a solution has already be discovered? Would putting the ADC right on the chip be some radically expensive redesign?"

Both Canon and Sony have patents for column-parallel ADC implementations. Canon filed for a Dual-Scale CP-ADC patent a couple of years ago, and quite probably had functional prototype (i.e. 120mp APS-H) sensors years before that (based on Canon's descriptions of the technology used in that sensor prototype.) Both Canon and Sony have patents for a variety of hardware level noise prevention concepts.

The real question is, why hasn't Canon employed technology they already own? Canon achieved high speed 9.5fps readout from a ONE HUNDRED AND TWENTY MEGAPIXEL APS-H sensor...YEARS ago. They have patents for backside illuminated five layer full frame sensors (as well as a host of additional patented innovations to improve the sensitivity and reduce light loss in such a sensor, something critical to a layered sensor's success against bayer designs). They have patents that should reduce dark current to obscenely low levels. They have demonstrated some of the most sensitive sensors on the planet. Why aren't they employing this amazing technology in their consumer products? While Sony, which has many similar patents for many very similar technologies, IS employing it all, and continues to employ new innovations as they are made? Same thing goes for Samsung...similar patents, similar technologies, and they are employing those technologies in consumer products.

That's the one thing about Canon I really don't understand. They aren't any innovative slouch...they are one of the most innovative companies in the world. They just don't seem to employ a LOT of the innovations they create. Canon sits on awesome technology, letting it rot away in some R&D corner of the company somewhere. I just hope 2015 is a year of radical change for Canon...I hope we SEE that layered BSI sensor in an actual product. If 2015 closes out without a hint of any truly ground breaking new sensor technology from Canon, I'll be pretty bummed. And confused...it makes no sense for a company to innovate and never actually employ the technology they own.

 

[Orangutan]

So then my next question would be "Can Canon not employ a similar design? Is such a design patented by Sony?" "If they can, what's stopping them from doing this if a solution has already be discovered? Would putting the ADC right on the chip be some radically expensive redesign?"

The real question is, why hasn't Canon employed technology they already own?

<snip>

Why aren't they employing this amazing technology in their consumer products? While Sony, which has many similar patents for many very similar technologies, IS employing it all, and continues to employ new innovations as they are made? Same thing goes for Samsung...similar patents, similar technologies, and they are employing those technologies in consumer products.

I think we can make a good guess at the answers to these questions: it's because they haven't had to. It costs money to re-tool for innovative new tech, and that cuts into both profits and R&D. Sony and Nikon have had to push out their best efforts to remain competitive; Samsung needs to put out their best to get into competition.

My hope (though not my rational expectation) is that Canon will push far enough ahead in their next retool to last for several iterations of the product cycle. Maybe that's what will be at work with the new high-MP camera: the megapixel count could be an opportunity to take it out of competition with the 1-series so they can put in the new sensor tech.

Let's hope Sony, Nikon and Samsung put more pressure on Canon.

Happy New Year to all. Go out and take some pictures!

 

[Lee Jay]

So, let's say the camera has a mode where the two halves of each pixel are read at different ISOs. For sake of argument, let's say ISO 100 and ISO 1600 are used.

So half the pixels have noise and DR quality of ISO 100 and the other half have a noise and DR quality from ISO 1600. How does that solve anything except give me less DR and more noise in half the pixels?

The ISO 1600 shots have less noise (by a lot) in the shadows than the ISO 100 shot taken at the same exposure. So the shadows are greatly less noisy that way.

 

[jrista]

So then my next question would be "Can Canon not employ a similar design? Is such a design patented by Sony?" "If they can, what's stopping them from doing this if a solution has already be discovered? Would putting the ADC right on the chip be some radically expensive redesign?"

The real question is, why hasn't Canon employed technology they already own?

<snip>

Why aren't they employing this amazing technology in their consumer products? While Sony, which has many similar patents for many very similar technologies, IS employing it all, and continues to employ new innovations as they are made? Same thing goes for Samsung...similar patents, similar technologies, and they are employing those technologies in consumer products.

I think we can make a good guess at the answers to these questions: it's because they haven't had to. It costs money to re-tool for innovative new tech, and that cuts into both profits and R&D. Sony and Nikon have had to push out their best efforts to remain competitive; Samsung needs to put out their best to get into competition.

My hope (though not my rational expectation) is that Canon will push far enough ahead in their next retool to last for several iterations of the product cycle. Maybe that's what will be at work with the new high-MP camera: the megapixel count could be an opportunity to take it out of competition with the 1-series so they can put in the new sensor tech.

Let's hope Sony, Nikon and Samsung put more pressure on Canon.

Happy New Year to all. Go out and take some pictures!

Your probably dead on. The bummer about Canon doing that is it pushes out any real competitive sensor IQ tech out another three years. So, if they just slide by again in 2015, maybe reemploying the evolutionary improvements in the 7D II in a high MP FF sensor...then it's probably going to be 2018 before we see a layered sensor. If they skip the layered sensor (or any other competitive new sensor tech) in 2018, it'll be 2021 before we see something truly new. Despite the lower dark current of the 7D II, it still "feels" like the last time Canon really did something amazing with their sensor technology was with the 5D II. I just hope I don't feel the same way in 2018 or 2021. Canon IS my preferred brand...but it's getting old "getting old" waiting for them to stop being...old.

 

[Famateur]

Such a silly comment is hardly worth responding to. Canon understands their market extremely well. There is ample evidence of that for anyone paying the least bit of attention.

Since a Canon Executive has just given an interview saying that they missed their customers need for a high megapix camera its fair to say Canon think's they actually got the market sentiment wrong on this count.

Also, Canon DSLR sales are not only plummeting but also doing so at a much faster rate than Canon expected only 12 months ago (leading to two downwards revisions of their estimated 2014 DSLR sales during this year).

Doubt the Canon board members are as impressed as you are.

DSLR sales are down across the industry -- it's not unique to Canon, nor is it a sign of losing competitive advantage. Rather, it's the combined impact of years of weak economies plus the ubiquity of smart phones equipped with pretty decent and ever-improving cameras. The point-and-shoot segment is evaporating, and with it revenue that would be driving more R&D, retooling, et cetera.

For a market leader, revenue drops like this move a company into more conservative strategies for long-term viability. Unless you're Sony (trying to make a splash or die trying) or Samsung (trying to enter the market as a legitimate competitor), market conditions like these are not favorable to parading out your best technology. You can gamble and go all-in with what you have, or you can keep an eye on the market and act when necessary.

Despite what goes on in forums, I suspect that DSLR sales across the market are not down because of pent-up demand of people waiting on the sidelines for a breakthrough in sensors from Canon. Why waste amazing technology on a generation of cameras that is not going to sell as well as they would with healthier economies around the world, especially if current models are selling well enough?

If Canon sensors suck so bad, why do I see so many white lenses on the sidelines of every football game I watch? Pretty sure they're not Sony lenses.

I know, I know -- landscape often demands more dynamic range. I get it. Perhaps, though, landscape is only a slice of the market Canon is supplying. : Weddings, portraits, wildlife, sports -- they seem to be doing pretty well with Canon gear. Canon need not panic over a few vocal photographers switching to Sony. Note that I said "panic" and not "pay attention". Canon does pay attention. When market share and long-term strategy requires it, Canon will bring out its tech.

I say, keep it coming, Sony and Samsung! Put the pressure on! In fact, everyone that complains about Canon sensor low ISO dynamic range, go buy a Sony! That's the best way to motivate Canon to blow us away in the next sensor generation...unless...there aren't enough of you to make a difference...