An 50.6mp EOS-1 Body Being Explored [CR2]

[jrista]

Is there anyone else who thinks that this is what should have been released *instead of* the 5ds?

Nope. The 1D series changed with the release of the 1Dx, and I suspect that if something materializes, it'll be a high resolution/high framerate camera that probably won't deliver the same kind of IQ as the 5Ds. Higher speed readout means more read noise, and optimizing for high ISO is obviously different than optimizing for low ISO. I think that is apparent in the design of the 5Ds and the A7s, both of which have interesting deviations from "normal" compared to other cameras from the same manufacturer (or using the same sensor technology). For example, most Exmor cameras have very low read noise at ISO 100, however the A7s has quite high read noise at low ISO, yet exceptional high ISO performance. The 5Ds has some of the best low ISO performance in the Canon lineup, but poorer high ISO performance (which I don't think is 100% related to pixel size, as I've seen better performance out of Nikon 24mp APS-C cameras at high ISO.)


The 5Ds has it's place. It also seems clear that is Canon's new studio line. I'm not sure what they could put into a 1D model that is better for studio, maybe one or two minor features like AF-point linked metering. I suspect the 1D line will remain the "blended" high performance line, and still geared towards the action shooter, while the 5Ds line will become the new studio line.

 

[jrista]

A Low Pass (LP) and Anti Aliasing Filter (AA) are the same thing.

Yep. My limited time with breadboards and signal processing in grad school taught me the same, but I've read that there were in fact two different layers to the stack with Nikon for the AA and some other LP filter, hence the convoluted story of:

D800 = LP + AA
D800E = LP
D810 = Nothing

But is Canon hedging on this because of drawbacks in removing (cancelling) the AA? Does defeating the AA effectively make the 5Dsr a stills-only camera? Is that the hang-up?

- A

Low pass and anti-aliasing filters are the same thing. The D800 had horizontal and vertical low pass filters, which are AA filters. These split the incoming high frequency spatial frequencies in the horizontal and vertical directions to blur them just enough to avoid interference patterns around the nyquist limit.


The D800E had only one, I forget which, let's say horizontal low pass filter. This was, for lack of a better term, a "positive" filter in that it split the high resolution spatial frequencies. Then it had another horizontal filter, only it was a "negative" filter in that it recombined the high resolution spatial frequencies, effectively canceling out the effect.


Both cameras also had IR cutoff filters to block near infrared light, and also attenuate red light to produce more "human sight" like reds (without that filter, reds become significantly stronger in DSLRs...one of the reasons astrophotographers like to remove these filters...lot of gas in space is hydrogen, which emits a lot of red light.)


The D810 doesn't have any low pass filters at all. It does have an IR cutoff filter, just not any low pass filters of any kind, "positive" or "negative".

 

[PureClassA]

Is there anyone else who thinks that this is what should have been released *instead of* the 5ds?

Nope. The 1D series changed with the release of the 1Dx, and I suspect that if something materializes, it'll be a high resolution/high framerate camera that probably won't deliver the same kind of IQ as the 5Ds. Higher speed readout means more read noise, and optimizing for high ISO is obviously different than optimizing for low ISO. I think that is apparent in the design of the 5Ds and the A7s, both of which have interesting deviations from "normal" compared to other cameras from the same manufacturer (or using the same sensor technology). For example, most Exmor cameras have very low read noise at ISO 100, however the A7s has quite high read noise at low ISO, yet exceptional high ISO performance. The 5Ds has some of the best low ISO performance in the Canon lineup, but poorer high ISO performance (which I don't think is 100% related to pixel size, as I've seen better performance out of Nikon 24mp APS-C cameras at high ISO.)


The 5Ds has it's place. It also seems clear that is Canon's new studio line. I'm not sure what they could put into a 1D model that is better for studio, maybe one or two minor features like AF-point linked metering. I suspect the 1D line will remain the "blended" high performance line, and still geared towards the action shooter, while the 5Ds line will become the new studio line.

Bingo. Thank you for putting it better than I could.

 

[ahsanford]

Low pass and anti-aliasing filters are the same thing. The D800 had horizontal and vertical low pass filters

[truncated]

...D810 doesn't have any low pass filters at all. It does have an IR cutoff filter, just not any low pass filters of any kind, "positive" or "negative".

Thanks. Very informative. Appreciated.

- A

 

[photonius]

Just curious about something...Is there any older Canon lens - like the EF100-400L - that will be able to resolve anything close to what a 50.6mp sensor can resolve, or will buying a 50.6mp camera require all new glass?

There are several old Canon lenses able to show much more detail with a 50 megapixel camera. A brief list:
<snip>
TS-E24mm F3.5L
TS-E17mm F4L
<snip>

The TS-E lenses will be key for those wanting maximum resolution and extended DOF. The diffraction limited aperture (DLA) is projected to be f/6.7, so f/16 or even f/11 won't be as usable at 50.6 as it is on the 5DII et al. Shooting with a T/S lens at f/5.6 or f/8 with tilt will be the way to give your image more DOF with a single exposure while keeping the aperture set for maximum sharpness.

I know some may chime in to explain all the technical details and limits here, but in practice, this is going to be the case. Hopefully this will push Canon to upgrade the non-L TS-E lenses (esp. the 45mm).

EDIT: Here's a good example from The-Digital-Picture:

300 f/2.8 IS II on the 1DsIII vs 7DII @ f/16

While not the most ideal example, the effect of diffraction on one of the sharpest lenses Canon makes is shocking on the 7DII (roughly same pixel pitch as 5Ds), even at f/11.

Good point.
I reprimanded often by people who say:
"more megapixel will never be worse" :-X

It depends if you are comparing apples to oranges, or not.

The 7DII crop is enlarged over twice as much so is an effective ff equivalent to f22. Change the crop camera aperture down 1 stop and you are looking at the same actual image qualities, or, dial f11 into the comparometer and keep the FF at f16, then you will see the true effects of diffraction at the same magnification and subject output, not so very different after all.

http://www.the-digital-picture.com/Reviews/ISO-12233-Sample-Crops.aspx?Lens=739&Camera=453&Sample=0&FLI=0&API=6&LensComp=739&CameraComp=963&SampleComp=0&FLIComp=0&APIComp=5

More MP are never worse, but look at more MP at 100% and they appear worse because you are magnifying them more. Magnify things the same and the more MP will always return more detail, even after DFA's. How much more is a test I am interested to carry out when I can come across a 5DSR, but that won't be soon and I don't expect to see much difference.

the crop should not be enlarged over 2x. the test chart should be shot at 1.6x distance, if tdp does it correctly.

Er, if you shoot something from 1.6 times the distance with the same lens and then reproduce it the same size you are enlarging it, and any IQ issues, diffraction, aberrations etc, more!

It is a 1.6 times linear enlargement, or a 1.6 x 1.6 area enlargement, which is 256% the area.

But that's not what should be happening with the test images at tdp. you have a standard test target. you use the same lens (e.g. 300mm). In one case on a FF body, let's say you are 3.5 meter away from the target to cover the test chart. With the maps-c body, you are 5.6 m away to cover the test chart. Now you zoom in 10x to show only the central circle of the test chart in both cases, the circle will be the same size. That's what you post on the web site. So, there is no extra enlargement.

Seriously?

Look, a 300mm lens has x magnification at x distance irrespective of what sensor is behind it, if you increase the distance the magnification gets less, if you reduce the distance the magnification gets higher.

How then can two cameras shot from different distances have the same size output? You enlarge the more distant shot more, that is what is being done in the linked tests. The only way you are comparing like for like with regards, in this instance, diffraction, is to shoot both cameras from the same distance with the same settings and enlarge them the same. But that is not what the first comparison does, my linked second comparison more closely emulates that from the results we have, but it still isn't quite the same.

well magnification is also Height-image divided height-object
And the two cameras have different sensor heights.
So, let's say, chart 1 meter high. M = 24/1000 for FF, for crop M = 15/1000
For tdp let's say we need to blow it up to 5 meters (and take then a 1/100 center crop).
FF: 5000/24 Total magnification from test chart to sensor to web site: 5000/24 * 24 / 1000 = 5
crop: 5000/15 total magnification 5000/15 * 15 /1000 = 5

 

[privatebydesign]

Just curious about something...Is there any older Canon lens - like the EF100-400L - that will be able to resolve anything close to what a 50.6mp sensor can resolve, or will buying a 50.6mp camera require all new glass?

There are several old Canon lenses able to show much more detail with a 50 megapixel camera. A brief list:
<snip>
TS-E24mm F3.5L
TS-E17mm F4L
<snip>

The TS-E lenses will be key for those wanting maximum resolution and extended DOF. The diffraction limited aperture (DLA) is projected to be f/6.7, so f/16 or even f/11 won't be as usable at 50.6 as it is on the 5DII et al. Shooting with a T/S lens at f/5.6 or f/8 with tilt will be the way to give your image more DOF with a single exposure while keeping the aperture set for maximum sharpness.

I know some may chime in to explain all the technical details and limits here, but in practice, this is going to be the case. Hopefully this will push Canon to upgrade the non-L TS-E lenses (esp. the 45mm).

EDIT: Here's a good example from The-Digital-Picture:

300 f/2.8 IS II on the 1DsIII vs 7DII @ f/16

While not the most ideal example, the effect of diffraction on one of the sharpest lenses Canon makes is shocking on the 7DII (roughly same pixel pitch as 5Ds), even at f/11.

Good point.
I reprimanded often by people who say:
"more megapixel will never be worse" :-X

It depends if you are comparing apples to oranges, or not.

The 7DII crop is enlarged over twice as much so is an effective ff equivalent to f22. Change the crop camera aperture down 1 stop and you are looking at the same actual image qualities, or, dial f11 into the comparometer and keep the FF at f16, then you will see the true effects of diffraction at the same magnification and subject output, not so very different after all.

http://www.the-digital-picture.com/Reviews/ISO-12233-Sample-Crops.aspx?Lens=739&Camera=453&Sample=0&FLI=0&API=6&LensComp=739&CameraComp=963&SampleComp=0&FLIComp=0&APIComp=5

More MP are never worse, but look at more MP at 100% and they appear worse because you are magnifying them more. Magnify things the same and the more MP will always return more detail, even after DFA's. How much more is a test I am interested to carry out when I can come across a 5DSR, but that won't be soon and I don't expect to see much difference.

the crop should not be enlarged over 2x. the test chart should be shot at 1.6x distance, if tdp does it correctly.

Er, if you shoot something from 1.6 times the distance with the same lens and then reproduce it the same size you are enlarging it, and any IQ issues, diffraction, aberrations etc, more!

It is a 1.6 times linear enlargement, or a 1.6 x 1.6 area enlargement, which is 256% the area.

But that's not what should be happening with the test images at tdp. you have a standard test target. you use the same lens (e.g. 300mm). In one case on a FF body, let's say you are 3.5 meter away from the target to cover the test chart. With the maps-c body, you are 5.6 m away to cover the test chart. Now you zoom in 10x to show only the central circle of the test chart in both cases, the circle will be the same size. That's what you post on the web site. So, there is no extra enlargement.

Seriously?

Look, a 300mm lens has x magnification at x distance irrespective of what sensor is behind it, if you increase the distance the magnification gets less, if you reduce the distance the magnification gets higher.

How then can two cameras shot from different distances have the same size output? You enlarge the more distant shot more, that is what is being done in the linked tests. The only way you are comparing like for like with regards, in this instance, diffraction, is to shoot both cameras from the same distance with the same settings and enlarge them the same. But that is not what the first comparison does, my linked second comparison more closely emulates that from the results we have, but it still isn't quite the same.

well magnification is also Height-image divided height-object
And the two cameras have different sensor heights.
So, let's say, chart 1 meter high. M = 24/1000 for FF, for crop M = 15/1000
For tdp let's say we need to blow it up to 5 meters (and take then a 1/100 center crop).
FF: 5000/24 Total magnification from test chart to sensor to web site: 5000/24 * 24 / 1000 = 5
crop: 5000/15 total magnification 5000/15 * 15 /1000 = 5

You keep making the same mistake, but this time you have confused yourself with maths in doing it.

Think if it like this. A lens projects an image onto the sensor plane, from any distance it does this the same size irrespective of what sensor size is behind it. If you change the distance the magnification changes.

If I use a ff camera and take a frame filling image of a test target (or anything) and then want the same framing from a Canon crop camera with the same lens I have to move back 1.6 times as far, in doing so that subject is projected onto the sensor 1.6 times linear smaller, or 40% the area. If I then want to reproduce those images the same size I have to magnify the crop camera image 1.6 times linear more, or 256% the area.

So I have made this little collage to help us along. The first two images are life sized FF and crop sensor images of a test target shot so the target fills the frame for both cameras, resize your browser window to make the FF image 36mm (or close to it) long.

These are what your sensor sees, and to make it so you need to be 1.6 times further away for the crop camera.

The second two images are enlargements of the first two such that we see a section of the test target the same size for both sensors. The FF crop is enlarged 5X, the crop camera crop has to be enlarged 8X (5x1.6=8) doing this makes the features in the two crops the same size.

The bottom two images are exactly what we are seeing in the linked tests. We can draw several conclusions from this:-[list type=decimal]
[*]The crop camera crop is enlarged more.
[*]The crop camera IQ is worse because it is enlarged more.
[*]Any aberrations are magnified more in the crop camera image.
[*]This is not a fair comparison of IQ.
[/list]

 

[weixing]

Just curious about something...Is there any older Canon lens - like the EF100-400L - that will be able to resolve anything close to what a 50.6mp sensor can resolve, or will buying a 50.6mp camera require all new glass?

There are several old Canon lenses able to show much more detail with a 50 megapixel camera. A brief list:
<snip>
TS-E24mm F3.5L
TS-E17mm F4L
<snip>

The TS-E lenses will be key for those wanting maximum resolution and extended DOF. The diffraction limited aperture (DLA) is projected to be f/6.7, so f/16 or even f/11 won't be as usable at 50.6 as it is on the 5DII et al. Shooting with a T/S lens at f/5.6 or f/8 with tilt will be the way to give your image more DOF with a single exposure while keeping the aperture set for maximum sharpness.

I know some may chime in to explain all the technical details and limits here, but in practice, this is going to be the case. Hopefully this will push Canon to upgrade the non-L TS-E lenses (esp. the 45mm).

EDIT: Here's a good example from The-Digital-Picture:

300 f/2.8 IS II on the 1DsIII vs 7DII @ f/16

While not the most ideal example, the effect of diffraction on one of the sharpest lenses Canon makes is shocking on the 7DII (roughly same pixel pitch as 5Ds), even at f/11.

Good point.
I reprimanded often by people who say:
"more megapixel will never be worse" :-X

It depends if you are comparing apples to oranges, or not.

The 7DII crop is enlarged over twice as much so is an effective ff equivalent to f22. Change the crop camera aperture down 1 stop and you are looking at the same actual image qualities, or, dial f11 into the comparometer and keep the FF at f16, then you will see the true effects of diffraction at the same magnification and subject output, not so very different after all.

http://www.the-digital-picture.com/Reviews/ISO-12233-Sample-Crops.aspx?Lens=739&Camera=453&Sample=0&FLI=0&API=6&LensComp=739&CameraComp=963&SampleComp=0&FLIComp=0&APIComp=5

More MP are never worse, but look at more MP at 100% and they appear worse because you are magnifying them more. Magnify things the same and the more MP will always return more detail, even after DFA's. How much more is a test I am interested to carry out when I can come across a 5DSR, but that won't be soon and I don't expect to see much difference.

the crop should not be enlarged over 2x. the test chart should be shot at 1.6x distance, if tdp does it correctly.

Er, if you shoot something from 1.6 times the distance with the same lens and then reproduce it the same size you are enlarging it, and any IQ issues, diffraction, aberrations etc, more!

It is a 1.6 times linear enlargement, or a 1.6 x 1.6 area enlargement, which is 256% the area.

But that's not what should be happening with the test images at tdp. you have a standard test target. you use the same lens (e.g. 300mm). In one case on a FF body, let's say you are 3.5 meter away from the target to cover the test chart. With the maps-c body, you are 5.6 m away to cover the test chart. Now you zoom in 10x to show only the central circle of the test chart in both cases, the circle will be the same size. That's what you post on the web site. So, there is no extra enlargement.

Seriously?

Look, a 300mm lens has x magnification at x distance irrespective of what sensor is behind it, if you increase the distance the magnification gets less, if you reduce the distance the magnification gets higher.

How then can two cameras shot from different distances have the same size output? You enlarge the more distant shot more, that is what is being done in the linked tests. The only way you are comparing like for like with regards, in this instance, diffraction, is to shoot both cameras from the same distance with the same settings and enlarge them the same. But that is not what the first comparison does, my linked second comparison more closely emulates that from the results we have, but it still isn't quite the same.

well magnification is also Height-image divided height-object
And the two cameras have different sensor heights.
So, let's say, chart 1 meter high. M = 24/1000 for FF, for crop M = 15/1000
For tdp let's say we need to blow it up to 5 meters (and take then a 1/100 center crop).
FF: 5000/24 Total magnification from test chart to sensor to web site: 5000/24 * 24 / 1000 = 5
crop: 5000/15 total magnification 5000/15 * 15 /1000 = 5

You keep making the same mistake, but this time you have confused yourself with maths in doing it.

Think if it like this. A lens projects an image onto the sensor plane, from any distance it does this the same size irrespective of what sensor size is behind it. If you change the distance the magnification changes.

If I use a ff camera and take a frame filling image of a test target (or anything) and then want the same framing from a Canon crop camera with the same lens I have to move back 1.6 times as far, in doing so that subject is projected onto the sensor 1.6 times linear smaller, or 40% the area. If I then want to reproduce those images the same size I have to magnify the crop camera image 1.6 times linear more, or 256% the area.

So I have made this little collage to help us along. The first two images are life sized FF and crop sensor images of a test target shot so the target fills the frame for both cameras, resize your browser window to make the FF image 36mm (or close to it) long.

These are what your sensor sees, and to make it so you need to be 1.6 times further away for the crop camera.

The second two images are enlargements of the first two such that we see a section of the test target the same size for both sensors. The FF crop is enlarged 5X, the crop camera crop has to be enlarged 8X (5x1.6=8) doing this makes the features in the two crops the same size.

The bottom two images are exactly what we are seeing in the linked tests. We can draw several conclusions from this:-[list type=decimal]
[*]The crop camera crop is enlarged more.
[*]The crop camera IQ is worse because it is enlarged more.
[*]Any aberrations are magnified more in the crop camera image.
[*]This is not a fair comparison of IQ.
[/list]

Hi,
Hmm... I think you are correct if both FF and crop had the same pixels density, but if both FF and crop had the same resolution (i.e. crop had higher pixel density than FF), than they should produce the similar quality image (under perfect condition).

Have a nice day.

 

[privatebydesign]

Just curious about something...Is there any older Canon lens - like the EF100-400L - that will be able to resolve anything close to what a 50.6mp sensor can resolve, or will buying a 50.6mp camera require all new glass?

There are several old Canon lenses able to show much more detail with a 50 megapixel camera. A brief list:
<snip>
TS-E24mm F3.5L
TS-E17mm F4L
<snip>

The TS-E lenses will be key for those wanting maximum resolution and extended DOF. The diffraction limited aperture (DLA) is projected to be f/6.7, so f/16 or even f/11 won't be as usable at 50.6 as it is on the 5DII et al. Shooting with a T/S lens at f/5.6 or f/8 with tilt will be the way to give your image more DOF with a single exposure while keeping the aperture set for maximum sharpness.

I know some may chime in to explain all the technical details and limits here, but in practice, this is going to be the case. Hopefully this will push Canon to upgrade the non-L TS-E lenses (esp. the 45mm).

EDIT: Here's a good example from The-Digital-Picture:

300 f/2.8 IS II on the 1DsIII vs 7DII @ f/16

While not the most ideal example, the effect of diffraction on one of the sharpest lenses Canon makes is shocking on the 7DII (roughly same pixel pitch as 5Ds), even at f/11.

Good point.
I reprimanded often by people who say:
"more megapixel will never be worse" :-X

It depends if you are comparing apples to oranges, or not.

The 7DII crop is enlarged over twice as much so is an effective ff equivalent to f22. Change the crop camera aperture down 1 stop and you are looking at the same actual image qualities, or, dial f11 into the comparometer and keep the FF at f16, then you will see the true effects of diffraction at the same magnification and subject output, not so very different after all.

http://www.the-digital-picture.com/Reviews/ISO-12233-Sample-Crops.aspx?Lens=739&Camera=453&Sample=0&FLI=0&API=6&LensComp=739&CameraComp=963&SampleComp=0&FLIComp=0&APIComp=5

More MP are never worse, but look at more MP at 100% and they appear worse because you are magnifying them more. Magnify things the same and the more MP will always return more detail, even after DFA's. How much more is a test I am interested to carry out when I can come across a 5DSR, but that won't be soon and I don't expect to see much difference.

the crop should not be enlarged over 2x. the test chart should be shot at 1.6x distance, if tdp does it correctly.

Er, if you shoot something from 1.6 times the distance with the same lens and then reproduce it the same size you are enlarging it, and any IQ issues, diffraction, aberrations etc, more!

It is a 1.6 times linear enlargement, or a 1.6 x 1.6 area enlargement, which is 256% the area.

But that's not what should be happening with the test images at tdp. you have a standard test target. you use the same lens (e.g. 300mm). In one case on a FF body, let's say you are 3.5 meter away from the target to cover the test chart. With the maps-c body, you are 5.6 m away to cover the test chart. Now you zoom in 10x to show only the central circle of the test chart in both cases, the circle will be the same size. That's what you post on the web site. So, there is no extra enlargement.

Seriously?

Look, a 300mm lens has x magnification at x distance irrespective of what sensor is behind it, if you increase the distance the magnification gets less, if you reduce the distance the magnification gets higher.

How then can two cameras shot from different distances have the same size output? You enlarge the more distant shot more, that is what is being done in the linked tests. The only way you are comparing like for like with regards, in this instance, diffraction, is to shoot both cameras from the same distance with the same settings and enlarge them the same. But that is not what the first comparison does, my linked second comparison more closely emulates that from the results we have, but it still isn't quite the same.

well magnification is also Height-image divided height-object
And the two cameras have different sensor heights.
So, let's say, chart 1 meter high. M = 24/1000 for FF, for crop M = 15/1000
For tdp let's say we need to blow it up to 5 meters (and take then a 1/100 center crop).
FF: 5000/24 Total magnification from test chart to sensor to web site: 5000/24 * 24 / 1000 = 5
crop: 5000/15 total magnification 5000/15 * 15 /1000 = 5

You keep making the same mistake, but this time you have confused yourself with maths in doing it.

Think if it like this. A lens projects an image onto the sensor plane, from any distance it does this the same size irrespective of what sensor size is behind it. If you change the distance the magnification changes.

If I use a ff camera and take a frame filling image of a test target (or anything) and then want the same framing from a Canon crop camera with the same lens I have to move back 1.6 times as far, in doing so that subject is projected onto the sensor 1.6 times linear smaller, or 40% the area. If I then want to reproduce those images the same size I have to magnify the crop camera image 1.6 times linear more, or 256% the area.

So I have made this little collage to help us along. The first two images are life sized FF and crop sensor images of a test target shot so the target fills the frame for both cameras, resize your browser window to make the FF image 36mm (or close to it) long.

These are what your sensor sees, and to make it so you need to be 1.6 times further away for the crop camera.

The second two images are enlargements of the first two such that we see a section of the test target the same size for both sensors. The FF crop is enlarged 5X, the crop camera crop has to be enlarged 8X (5x1.6=8) doing this makes the features in the two crops the same size.

The bottom two images are exactly what we are seeing in the linked tests. We can draw several conclusions from this:-[list type=decimal]
[*]The crop camera crop is enlarged more.
[*]The crop camera IQ is worse because it is enlarged more.
[*]Any aberrations are magnified more in the crop camera image.
[*]This is not a fair comparison of IQ.
[/list]

Hi,
Hmm... I think you are correct if both FF and crop had the same pixels density, but if both FF and crop had the same resolution (i.e. crop had higher pixel density than FF), than they should produce the similar quality image (under perfect condition).

Have a nice day.

AAAAAAAHHHHHHHHHH!!!!!!!!! :

No, they should not. Pixel density has got absolutely nothing to do with anything here, we are talking enlargement, or, magnification, pixel density has got zero impact on that in any modern digital camera. Whilst you are having a nice day go read this http://www.josephjamesphotography.com/equivalence/ then get back to me with a comment that actually makes sense.

 

[jrista]

Hi,
Hmm... I think you are correct if both FF and crop had the same pixels density, but if both FF and crop had the same resolution (i.e. crop had higher pixel density than FF), than they should produce the similar quality image (under perfect condition).

Have a nice day.

That sounds nice and neat, but it is not the case. Higher pixel density can affect final image resolution...but only in reach-limited situations. It does not actually mean IQ is the same if sensor sizes are different. It's because the SNR of the smaller sensor is lower than the SNR of the larger sensor.

Let's just assume we have two cameras with 20 megapixels, one is 36x24mm and one is 18x12mm. This is an area difference of 4x, so the larger sensor has pixels 4x as large. As such, let's say the pixel FWC of the smaller is 25,000e-, and the pixel FWC of the larger is 100,000e-. You expose both cameras off an 18% gray card, so you get a perfect middle tone.

Your two photos are technically the same...you exposed both the same way on an 18% gray field, and both images have exposures exactly half their FWC. That means the average level in each pixel of the smaller sensor is 12,500e-, and in each pixel of the larger sensor is 50,000e-. The difference in the number of electrons per pixel (the signal) should already be telling you something here, but let's figure out the SNR for each just to be sure, and for now let's assume there is no read noise:

largeSNR = 50,000/SQRT(50,000) = 223.6:1
smallSNR = 12,500/SQRT(12,500) = 111.8:1

The larger sensor clearly has the higher SNR. Higher SNR is what leads to better IQ. We can convert this into decibels, which would then correlate with many test site data:

largeSNRdB = 20 * log(223.6) = 47dB
smallSNRdB = 20 * log(111.8) = 41dB

In decibels, a difference of 6dB is equivalent to a stop improvement. So, the larger sensor here gains exactly a 1-stop improvement in terms of IQ over the smaller sensor. The smaller sensor is exactly 1/4 the size of the larger sensor, or a difference of 4x. Image quality scales as the square root of that, or 2x...a factor of two, a doubling or halving.

In terms of noise, the larger sensor wins out here because each and every one of it's 20 million pixels are gathering more light. Assuming you were identically framed, there is no difference in final resolution either...you put the same amount of pixels onto the same area of your gray card in this case, but if it was a bird, or a mountain, or a person's portrait, it would be the same. Both cameras have 20 megapixels, but one has pixels 4x as large, and therefor it gathers more light for the same subject.

 

[photonius]

Just curious about something...Is there any older Canon lens - like the EF100-400L - that will be able to resolve anything close to what a 50.6mp sensor can resolve, or will buying a 50.6mp camera require all new glass?

There are several old Canon lenses able to show much more detail with a 50 megapixel camera. A brief list:
<snip>
TS-E24mm F3.5L
TS-E17mm F4L
<snip>

The TS-E lenses will be key for those wanting maximum resolution and extended DOF. The diffraction limited aperture (DLA) is projected to be f/6.7, so f/16 or even f/11 won't be as usable at 50.6 as it is on the 5DII et al. Shooting with a T/S lens at f/5.6 or f/8 with tilt will be the way to give your image more DOF with a single exposure while keeping the aperture set for maximum sharpness.

I know some may chime in to explain all the technical details and limits here, but in practice, this is going to be the case. Hopefully this will push Canon to upgrade the non-L TS-E lenses (esp. the 45mm).

EDIT: Here's a good example from The-Digital-Picture:

300 f/2.8 IS II on the 1DsIII vs 7DII @ f/16

While not the most ideal example, the effect of diffraction on one of the sharpest lenses Canon makes is shocking on the 7DII (roughly same pixel pitch as 5Ds), even at f/11.

Good point.
I reprimanded often by people who say:
"more megapixel will never be worse" :-X

It depends if you are comparing apples to oranges, or not.

The 7DII crop is enlarged over twice as much so is an effective ff equivalent to f22. Change the crop camera aperture down 1 stop and you are looking at the same actual image qualities, or, dial f11 into the comparometer and keep the FF at f16, then you will see the true effects of diffraction at the same magnification and subject output, not so very different after all.

http://www.the-digital-picture.com/Reviews/ISO-12233-Sample-Crops.aspx?Lens=739&Camera=453&Sample=0&FLI=0&API=6&LensComp=739&CameraComp=963&SampleComp=0&FLIComp=0&APIComp=5

More MP are never worse, but look at more MP at 100% and they appear worse because you are magnifying them more. Magnify things the same and the more MP will always return more detail, even after DFA's. How much more is a test I am interested to carry out when I can come across a 5DSR, but that won't be soon and I don't expect to see much difference.

the crop should not be enlarged over 2x. the test chart should be shot at 1.6x distance, if tdp does it correctly.

Er, if you shoot something from 1.6 times the distance with the same lens and then reproduce it the same size you are enlarging it, and any IQ issues, diffraction, aberrations etc, more!

It is a 1.6 times linear enlargement, or a 1.6 x 1.6 area enlargement, which is 256% the area.

But that's not what should be happening with the test images at tdp. you have a standard test target. you use the same lens (e.g. 300mm). In one case on a FF body, let's say you are 3.5 meter away from the target to cover the test chart. With the maps-c body, you are 5.6 m away to cover the test chart. Now you zoom in 10x to show only the central circle of the test chart in both cases, the circle will be the same size. That's what you post on the web site. So, there is no extra enlargement.

Seriously?

Look, a 300mm lens has x magnification at x distance irrespective of what sensor is behind it, if you increase the distance the magnification gets less, if you reduce the distance the magnification gets higher.

How then can two cameras shot from different distances have the same size output? You enlarge the more distant shot more, that is what is being done in the linked tests. The only way you are comparing like for like with regards, in this instance, diffraction, is to shoot both cameras from the same distance with the same settings and enlarge them the same. But that is not what the first comparison does, my linked second comparison more closely emulates that from the results we have, but it still isn't quite the same.

well magnification is also Height-image divided height-object
And the two cameras have different sensor heights.
So, let's say, chart 1 meter high. M = 24/1000 for FF, for crop M = 15/1000
For tdp let's say we need to blow it up to 5 meters (and take then a 1/100 center crop).
FF: 5000/24 Total magnification from test chart to sensor to web site: 5000/24 * 24 / 1000 = 5
crop: 5000/15 total magnification 5000/15 * 15 /1000 = 5

You keep making the same mistake, but this time you have confused yourself with maths in doing it.

Think if it like this. A lens projects an image onto the sensor plane, from any distance it does this the same size irrespective of what sensor size is behind it. If you change the distance the magnification changes.

If I use a ff camera and take a frame filling image of a test target (or anything) and then want the same framing from a Canon crop camera with the same lens I have to move back 1.6 times as far, in doing so that subject is projected onto the sensor 1.6 times linear smaller, or 40% the area. If I then want to reproduce those images the same size I have to magnify the crop camera image 1.6 times linear more, or 256% the area.

So I have made this little collage to help us along. The first two images are life sized FF and crop sensor images of a test target shot so the target fills the frame for both cameras, resize your browser window to make the FF image 36mm (or close to it) long.

These are what your sensor sees, and to make it so you need to be 1.6 times further away for the crop camera.

The second two images are enlargements of the first two such that we see a section of the test target the same size for both sensors. The FF crop is enlarged 5X, the crop camera crop has to be enlarged 8X (5x1.6=8) doing this makes the features in the two crops the same size.

The bottom two images are exactly what we are seeing in the linked tests. We can draw several conclusions from this:-[list type=decimal]
[*]The crop camera crop is enlarged more.
[*]The crop camera IQ is worse because it is enlarged more.
[*]Any aberrations are magnified more in the crop camera image.
[*]This is not a fair comparison of IQ.
[/list]

I have not confused myself, my math is exactly what you illustrate with your sample images. The only thing you left out is that the magnification is different to project the test chart onto the different sized sensors.

"The crop camera IQ is worse because it is enlarged more." this statement is only true under some conditions. If you have a 8MP FF sensor, and a 24 MP crop sensor, the crop will be better (i.e opposite what you show above).

"This is not a fair comparison". That is not the point. The web site simply provides images of lenses on different sensors, all magnified to the same extent compared to the original chart. That allows you to get
an approximate idea how good the image is of that lens/body combo compared to some other lens combo.
And yes, the crop sensor will show CA etc. 1.6x (linear) more than FF.

 

[tiredofstitching]

Hey, wouldn't you folks trim the quotes to what you really wish to answer? It would make reading the posts so much more enjoyable!

8)

 

[StudentOfLight]

Hey, wouldn't you folks trim the quotes to what you really wish to answer? It would make reading the posts so much more enjoyable!

8)

+1 8)

 

[privatebydesign]

Hey, wouldn't you folks trim the quotes to what you really wish to answer? It would make reading the posts so much more enjoyable!

8)

+1 8)

For two reasons, first, it maintains context, I hate it when people partial quote and deliberately completely misrepresent what you actually wrote, just like photonius quoting my my partial comment "This is not a fair comparison" then ripping me apart in saying that isn't the point of the website, I know that! But it was the point of the initial linked comparison about which we are talking and that the poster has subsequently agreed is not correct and not a fair comparison. And second, I kinda like to see how far the nesting will go ;D it seems like they go into a black hole of quotes!

Also, I gave up with photonius, I expect most other people can see the error of using those two samples as a comparison of diffraction, all those that don't have their head up their butts and confuse pixel density with enlargement anyway.