Canon's 120mp APS-H Sensor

[kubelik]

pierlux, I think scalesusa was referring to my question about DLA; like I said, I don't fully understand it and would not be averse to some schooling in light physics.

I like where you're taking this train of thought with the pixel binning down to a 30 MP sensor, and I think chrome_dude has some interesting thoughts in terms of pixels with different ranges of sensitivity even in the same color channel. it makes this bit of news seems a lot less like canon losing its marbles and more like an earnest scientific (and economic) endeavor.

I am amazed that they're calculating 9.5 fps with that number of pixels; is the pixel binning occurring on-sensor to allow this kind of throughput?

 

[rugged]

Some articles about this sensor say we won't see it in our lifetime. If that's true, why would they bother making it? Or perhaps they are implying that the cost and storage media is the bottleneck?

 

[Grendel]

I doubt we will see it anytime soon -- one of the foot notes in the original news release says:

In 2007, the company successfully developed an APS-H-size sensor with approximately 50 million pixels.

Haven't seen that one anywhere yet..

 

[unfocused]

I guess I better get a bigger memory card... and a bigger hard drive.

Seriously though, as some others have suggested this could be a real game-changer if it gets implemented. Think of what it would mean for sports and wildlife shooters. Use a 200mm f2.8 lens and then crop the image for the equivalent of an 800mm (or longer).

I'm wondering what the Canon lens division thinks of this. Would the "Big White Lenses" become obsolete?

Would sports shooters need to tether an auxiliary memory drive to their cameras, comparable to the bulk film magazines they used to use with motor drives?

Imagine the fits this would give professional sports teams if any fan could shoot high quality action shots from the stands.

No doubt there are a lot of obstacles standing in the way of implementation. Complementary technologies (memory, lenses, etc. etc.) need to catch up. So, we probably won't see this for several years. But, clearly the march to ever-increasing resolution is accelerating.

 

[Lee Jay]

I know I'm not nearly good enough with the physics to know this for sure, but wouldn't a 120 MP APS-H have a ridiculously low bar for lens diffraction?

Adding pixels to the same sensor size and lens combination can never, ever make diffraction worse. In fact, it makes it better.

There are certainly diminishing returns to adding pixels as diffraction takes its bite out of your resolution, but in my own tests of where the point of no further visual benefit of more pixels is located, the 7D's sensor is capable of extracting all the detail visually available in an f/20 lens. That means pixels could get a heck of a lot smaller than even those tiny 7D pixels before you'd be sucking the life out of an f/2.8 supertele. This is consistent with astrophotography experience where people routinely use extremely slow f-numbers to extract all the detail they can get from planetary images. I was surprised (shocked, really) at how much more detail I could get at 3800mm and f/30 compared with 1900mm and f/15 even with 40D-sized pixels. Have a look for yourself: http://photos.imageevent.com/sipphoto/samplepictures/Jupiter%20f30%20versus%20f15%20comparison.jpg

That's f/30 on the left. While you might have thought 40D-sized pixels were already diffraction-limited at f/15, obviously going to f/30 still extracted a lot more detail. That switch is the same as making the pixels half as big.

 

[Inst]

It's probably just a technology demonstrator sensor demonstrating that their techs can scale up the S10/S11 technology to APS-H levels. Actually mass-producing such a device would most likely be outlandishly expensive with current fabrication processes; I wouldn't be surprised to see $500 or above per sensor. Also, can you imagine the MTF needed on lenses for this system? You'd be lens-limited until Canon could hire Carl Zeiss off Sony to produce $10,000 lenses. And how about post-processing the damn thing? You will be dealing with 120 MP raw files. You're probably already shelling out 4 digits for EIZO LCDs, are you looking forward to buying dual GTX480 ($500 or $600 a pop) GPUs to fuel CUDA-based Photoshop CS5, along with a $1000 CPU and probably 16 gigs of ram?

I think the only way to bring this to market would be to market it as a ridiculous sensor toy in an MF-like modular camera; you'd get upgradeable DIGIC processors starting at 1 fps, and then later on as processing improves you'd move up to 8 or 10 fps. Your lenses would be completely inadequate, but if you own a very high-end outfit I suppose you could wow clientele with your $100,000 wanna-be medium format.

Dynamic range does have an inverse correlation to pixel size, but it's not as bad as people make it out to be; 120 MP would have more than 6.6 times the signal to noise ratio than the 7D based on megapixel increases while ignoring the effects of the smaller sensor size.

Ehhh...

Looking at the G11's dxomark data, it would have a 3rd of a stop less full-picture performance or identical performance at ISO 3200 than the existing 1D4, depending on how you treat manufacturer ISOs and between one stop and five thirds of a stop less full-picture performance than the D3s, depending on whether you consider the fact that the D3s's manufacturer ISO 3200 is somewhat below the G11's. In Dynamic Range, using SnR comparisons, which is typically valid for "usable dynamic range", but not total dynamic range, the 120 MP super G11 would scale to about 14 EV, which is above the D3x, I believe.

DLA for the 120mp APS-H should kick in around f/4, I believe, so everyone would be advised to use f/2.8 lenses, further adding to the cost.

Everything considered, I would be willing to buy a 120mp APS-H provided that I had the million dollars needed to pay for the camera, the lens, the computer, and nothing better to spend it on. Even if I were severely lens-limited, compared to the existing cameras, think of the ISO 50 resolution on this baby! It would be able to resolve at least twice the MTF detail over existing devices, and when sensors do catch up you'd be able to do micro-biology with macro lenses.

 

[Sebastian]

I know I'm not nearly good enough with the physics to know this for sure, but wouldn't a 120 MP APS-H have a ridiculously low bar for lens diffraction?

Adding pixels to the same sensor size and lens combination can never, ever make diffraction worse. In fact, it makes it better.

No. Diffraction is a property solely of the lens (and the aperture therein). Period.

On the other hand, the sensor's pixel pitch only affects the circle of confusion. (No, that's not what some guys here seem to be in. )

Of course, a smaller pixel pitch leads to a smaller circle of confusion, thus demanding a higher quality of the lens to get what appears to be a "sharp" image.


HTH,

Sebastian

 

[Grendel]

Here's a good read for backgropund info on diffraction. Bottom of the page has a link to part 2.

 

[Peerke]

DLA for the 120mp APS-H should kick in around f/4, I believe, so everyone would be advised to use f/2.8 lenses, further adding to the cost.

If you pixel peep that is, but who want to view a 120 MP picture at 100% on his monitor?

 

[328somewhere]

Here is a small theory.
What about the use in medium format applications. Wasn't it just rumored that canon was looking for a medium format company to purchase? I am no camera guru but wouldn't there be enough room in a medium format body to use this sensor to it's potential? Call it some kind of crossover body. With an ef mount on a medium format body they could take over the world. Maybe this is just a hint but you don't make an announcement like this without a reason.
Opinions?

 

[pierlux]

Here is a small theory.
What about the use in medium format applications. Wasn't it just rumored that canon was looking for a medium format company to purchase? I am no camera guru but wouldn't there be enough room in a medium format body to use this sensor to it's potential? Call it some kind of crossover body. With an ef mount on a medium format body they could take over the world. Maybe this is just a hint but you don't make an announcement like this without a reason.
Opinions?

Format, with respect to photography, is relative to sensor (or film) size, not the body. This is an APS-H sized sensor, you can put it in whatever sized camera, but the format will always be APS-H. Medium format refers to cameras which produce 45x60mm, 60x60mm or 60x90mm images in the case of film, or larger than 24x36mm in the case of digital cameras.

 

[pierlux]

There must be good reasons why Canon made this announcement.

It's probably just a technology demonstrator sensor demonstrating that their techs can scale up the S10/S11 technology to APS-H levels.

why would they bother making it?

Maybe this announcement was made because of this:
http://nikonrumors.com/2010/08/20/nikon-now-producing-their-own-sensors.aspx#more-13210

Probably Canon is just stating: "Whatever you do, we are always one million lightyears ahead of you in sensor technology!" And if this is true, I have to reconsider all I have said in my previous post... :-[

BTW (and off topic), I've been always wondeing why Nikon, being the world's No. 2 stepper maker ( http://www.nikonprecision.com/ ), does not produce its own sensors instead of buying them from Sony. Marketing misteries...

 

[gkreis]

Now, it is true that a higher resolution sensor can see the diffraction better, just as it provides better resolution of the subject, but a sensor does not affect diffraction.

A 120mp APS-H sensor has about the same pixel density as the 12 mp G11 / S90. The lenses seem to work fine on it, and are not as wide a aperture as are available on DSLR's.


120mp/5.19 cm square = 23.1 mp/square cm, approximately the same as the 12mp G11.

Of course the sensor doesn't cause diffraction, it just records it.

Why doesn't a G11 have f stops larger than f8?

 

[gkreis]

Seriously though, as some others have suggested this could be a real game-changer if it gets implemented. Think of what it would mean for sports and wildlife shooters. Use a 200mm f2.8 lens and then crop the image for the equivalent of an 800mm (or longer).

I think the biggest problem is micro blur. If seem to recall reading a paper from Canon that addressed the issue of 7D type density having issues with micro blur as a source of softness. Anyone remember it?

Basically, the idea is that if there are more pixels in the same area, they are better able to register movement during taking of the image. So if you crop way down, you might be get a blurrier image than you expect unless you had an extremely fast shutter speed. So hopes of heavily cropping these super dense sensors might be dashed in real-life, when those minor focus issues are 'blown up'.

 

[Lee Jay]

I know I'm not nearly good enough with the physics to know this for sure, but wouldn't a 120 MP APS-H have a ridiculously low bar for lens diffraction?

Adding pixels to the same sensor size and lens combination can never, ever make diffraction worse. In fact, it makes it better.

No. Diffraction is a property solely of the lens (and the aperture therein). Period.

On the other hand, the sensor's pixel pitch only affects the circle of confusion. (No, that's not what some guys here seem to be in. )

Of course, a smaller pixel pitch leads to a smaller circle of confusion, thus demanding a higher quality of the lens to get what appears to be a "sharp" image.


HTH,

Sebastian

Even if you are diffraction-limited, adding pixels will lead to a sharper image with more detail. This is because the blur caused by diffraction and the blur caused by the sampling add together (geometrically). That's why I said what I said. I didn't really mean to say the diffraction itself would improve, just that the total blur would be reduced.

Likewise, your last sentence is just wrong. More pixels means a more detailed image from the same lens at the same final viewing size. Now, if you meant at 100%, that's just not a realistic way to compare when you're changing the amount of enlargement between the two image, which is what you are doing by looking at two images with different pixel sizes both at 100%.

I can prove this with image samples, if you'd like.

 

[NormanBates]

I can prove this with image samples, if you'd like.

with something similar to this, I guess:
http://www.similaar.com/foto/mpix/mpix.html

in any case, I can think of a lot of reasons Canon has to manufacture a limited number of these sensors, for example:
* to research data collection and data manipulation processes associated with a high-mpix-count chip
* to research the effects of increased pixel density in DSLR sensors (how does it affect noise and DR? how can we process the data so that they improve with megapixel-density? how dense can we get before diffraction precludes any further detail improvements?)
* to research novel image processing methods, like those proposed by some forum members here in terms of diffraction-control, in-sensor-IS, etc

it's a research chip, just like intel's 48-core microprocessor

 

[kufucius]

I can prove this with image samples, if you'd like.

with something similar to this, I guess:
http://www.similaar.com/foto/mpix/mpix.html

in any case, I can think of a lot of reasons Canon has to manufacture a limited number of these sensors, for example:
* to research data collection and data manipulation processes associated with a high-mpix-count chip
* to research the effects of increased pixel density in DSLR sensors (how does it affect noise and DR? how can we process the data so that they improve with megapixel-density? how dense can we get before diffraction precludes any further detail improvements?)
* to research novel image processing methods, like those proposed by some forum members here in terms of diffraction-control, in-sensor-IS, etc

it's a research chip, just like intel's 48-core microprocessor

In that case, I wonder why they don't do it APS-C or Full frame? Weird

 

[NormanBates]

surely there's a reason, but we can only speculate

someone said here it could be because making a bigger chip would require special tools, but, to my understanding, they already went over that line with APS-H: many standard CMOS-chip-making tools can only be used for making chips under 24mm in their biggest dimension, which, here, means APS-C