On Landscape Review the Canon 5DS/R

[Josh-H]

Very thorough and interesting reading. Some pretty impressive performance from recent Canon optics.
On Landscape Review with imatest

[Ray]

It's good to see such a detailed review. Thanks for the link, Josh.

There are many folks who probably think they don't need 50mp but sometimes those same people will spend money to upgrade to a sharper and more expensive lens. This article really puts into perspective the benefits of the extra resolution of the 5DS(R). A 5DSR image downsampled to the same size as a 5D2 or 5D3 image really is sharper, just as an expensive prime is sharper than a zoom.

One of the more revealing comparisons in the article was between the 5D3 at F16 and the 5DS at F22. The detail is about the same. I'm reminded of my own comparisons years ago with the original 5D. I found very little fall-off between F11 and F16, but dropping down to F22 resulted in a noticeably softer image which was generally unacceptable.

Now for the big surprise. This extra detail carried on all the way to f/22, so much so that the f/22 file had as much detail as the f/16 file. i.e. you can use a stop smaller aperture and hence get more depth of field for your money. - See more at: https://www.onlandscape.co.uk/2015/07/testing-the-canon-5dsr/#sthash.rpuCroIg.dpuf 

[NancyP]

Thanks. This is a very interesting article to me - I am a Canonista currently happily shooting with a 6D at 20 MP - approximately equivalent to the 5D3, but less pattern noise. The bit about getting an additional f/stop worth of sharpness by downsizing is something to think about, for landscape use, at least if not shooting with a T/S lens. But I want a TS-E lens too! In the real world, what I NEED (relatively speaking) is a computer replacement (currently using 2010 non-Retina MacBookPro) before I consider a high-pixel camera.

[jeffreybehr]

...I am a Canonista currently happily shooting with a 6D at 20 MP--approximately equivalent to the 5D3 but with less pattern noise. The bit about getting an additional f/stop worth of sharpness by downsizing is something to think about, for landscape use, at least if not shooting with a T/S lens. But I want a TS-E lens too! In the real world, what I NEED (relatively speaking) is a computer replacement (currently using 2010 non-Retina MacBookPro) before I consider a high-pixel camera.

New T/S (and other?) lens, new computer (and printer, too?), new 50MP body...hi-end fotografy, just like hi-end audio, is an expensive proposition, huh?!?!?!

I call myself a Canonite.   

[bjanes]

It's good to see such a detailed review. Thanks for the link, Josh.

There are many folks who probably think they don't need 50mp but sometimes those same people will spend money to upgrade to a sharper and more expensive lens. This article really puts into perspective the benefits of the extra resolution of the 5DS(R). A 5DSR image downsampled to the same size as a 5D2 or 5D3 image really is sharper, just as an expensive prime is sharper than a zoom.

One of the more revealing comparisons in the article was between the 5D3 at F16 and the 5DS at F22. The detail is about the same. I'm reminded of my own comparisons years ago with the original 5D. I found very little fall-off between F11 and F16, but dropping down to F22 resulted in a noticeably softer image which was generally unacceptable.

Tim Parkin has an impressive testing setup including a wall full of Imatest targets and a Stack Shot (motorized macro focusing rail) to allow very precise focus bracketing and he obviously knows what he is doing, but I have a few comments that could invite further discussion.

He rates resolution at MTF of 5 or 9%, somewhat between the Dawes and Rayliegh limits, and notes that contrast is very low at these resolutions but can be increased by proper sharpening. Deconvolution algorithms are generally regarded as the best sharpening to mitigate diffraction blurring. He used Lightroom sharpening with the detail very low and a 0.6 px radius. For f/22, that radius is probably too small as judged by testing with Bart van der Wolf's slanted edge tool (which no longer appears to be available). Lightroom sharpening does use some type of deconvolution algorithm that maximizes deconvolution with the detail slider all the way to the right (large amount of detail). With this sharpening, he obtains resolutions beyond the Nyquist limit and I'm not certain if he is measuring true or false detail (sharpening artifacts).

In any event, I don't think his example of rocks or whatever is the best image to judge resolution--it has relatively little high frequency content, and I doubt that his sharpening can reverse the effects of diffraction at f/22. As an example, here are some results with the Nikon D800e and the Zeiss 135/f2.0 Apo lens using Bart's sinusoidal Siemen's star. The black circle indicates the Nyquist limit of the sensor, 103 cy/mm. At the optimum aperture of f/4 and without sharpening, the lens can resolve very close to Nyquist as shown here.



Focus Magic gives much improved contrast at f/4



At f/22, resolution is 72 cy/mm, well short of Nyquist. The sinusoidal bars do not extend to the Nyquist resolution as shown. Note that contrast is low throughout the target.



Here is the result of deconvolution sharpening with FocusMagic. Note contrast is improved, but resolution is little affected.



Here is f/16 with Focus Magic sharpening:



In summary, I would say that one should avoid f/22, if possible. You can restore contrast, but not resolution.

Regards,

Bill

[Ray]

Bill,
Convolution sharpening techniques are a separate issue. As you know, DXOMark, for example, do not convert the RAW images prior to taking their measurements because the quality and characteristics of the converter would cause a bias.

I'd say if one has raised one's resolution standards after getting used to the increased resolution provided by the 5DS(R), then one might be disappointed when comparing an F22 shot with an F16 shot using the same camera.

However, if one has frequently used F16 with a 5D3 and found the resolution quite acceptable for reasonably large prints, but has avoided using F22 with the 5D3 because the results are noticeably soft, then the 5DS(R) at F22 should produce more satisfying results than the 5D3 at F16, when the image is downsampled to the 5D3 file size, and whenever maximum DoF is desired of course.

I have assumed that the camparison crops shown in the review are of an area that is in focus. It would be interesting to see how much more detail is visible in the 5DS shot (at F22) in areas of the scene furthest away from the plane of focus.

I don't see any problem using a textured surface for such comparisons. This relates more closely to the nature of 'real-world' examples that most people encounter in landscapes.

I recall a few years ago comparing the last Canon camera I bought, the 15mp 50D, with the previous model, the 10mp 40D, photographing a banknote fixed to the wall. It was interesting that the 50D at F11 was about as sharp as the 40D at F8, using the same high quality prime lens and downsampling to 10mp. The 50D at F16 was also as detailed as the 40D at F11, after downsampling. However the 50D at F22 (if my memory serves me) was not quite as detailed as the 40D at F16. It was very close, but pixel-peeping at 100% and even 200%, I had to admit that the 40D at F16 had a very slight edge.

Now the 50D has only a 50% increase in pixel count over the 40D. The 5DS has over a 100% increase in pixel count over the 5D3. I can believe that the 5DS image at F22 really does have as much detail as the 5D3 at F16.

[Jack Hogan]

I am with Bill: nice review but despite the impressive setup the linear spatial resolution measurements are hoaky in the context of hardware evaluation/comparison.

Introducing (especially non-standard, non open source) sharpening into such tests immediately turns a fine quantitative evaluation of the MTF of the hardware (supposedly providing the same physical results independently of testing methodology) into a qualitative exercise on the artistic drawing capabilities of the post processing package used -  retaining only a tenuous link to hw performance, if at all. Might as well outline every edge in the image with black and white indelible markers and measure their MTF.   Here and here is a bit more on this subject.

Jack

[bjanes]

I am with Bill: nice review but despite the impressive setup the linear spatial resolution measurements are hoaky in the context of hardware evaluation/comparison.

Introducing (especially non-standard, non open source) sharpening into such tests immediately turns a fine quantitative evaluation of the MTF of the hardware (supposedly providing the same physical results independently of testing methodology) into a qualitative exercise on the artistic drawing capabilities of the post processing package used -  retaining only a tenuous link to hw performance, if at all. Might as well outline every edge in the image with black and white indelible markers and measure their MTF.   Here and here is a bit more on this subject.

Jack

Jack,

Thanks for your input and the additional information provided in your linked articles. As Ray pointed out, when one is comparing various camera systems, it makes no sense to use sharpening, which introduces more variables into the situation. As you point out in the links, it may be desirable to eliminate processing altogether and look directly at the raw file. An intermediate approach would be to use dcraw with neutral parameters and linear output. This option is offered by Imatest.

However, when one is evaluating a given camera/lens combination, rendering the image with reasonable sharpening may be advisable, since this would reproduce the workflow used in practical photography with that combination. For example, I might want to find the optimum aperture and the effects of diffraction at smaller apertures an the extent that one can recover MTF.

Cheers,

Bill

[Ray]

However, when one is evaluating a given camera/lens combination, rendering the image with reasonable sharpening may be advisable, since this would reproduce the workflow used in practical photography with that combination. For example, I might want to find the optimum aperture and the effects of diffraction at smaller apertures an the extent that one can recover MTF.

I agree. There must now be quite a few people who own both a 5D3 and the new 5DS(R). Here's the opportunity for them to present their own tests comparing the 5DS at F22 with the 5D3 at F16. Don't by shy! 

[Jack Hogan]

However, when one is evaluating a given camera/lens combination, rendering the image with reasonable sharpening may be advisable, since this would reproduce the workflow used in practical photography with that combination. For example, I might want to find the optimum aperture and the effects of diffraction at smaller apertures an the extent that one can recover MTF.

All agreed that that is indeed a worthwhile exercise - albeit quite different from the quantitative one related to a linear 'shift invariant' imaging system providing a response dictated by physics.  It is instead a perceptual exercise evaluated qualitatively and subjectively, which will give different results depending on the software, the subject and the user.  My main beef with the article is using MTF curves in this latter context: they give the impression of a level of precision that is in fact not there.  One can pretty well make them come out as one pleases by playing with USM, local contrast and deconvolution sliders during conversion or in post (e.g. LR/ACR).  So if one can make them come out as one pleases, what's their use in a hardware comparison?  To make the results seem more 'scientific'?  They are no better than simply looking at the compromises off the pretty pictures themselves and deciding which appears to be more pleasingly sharp.  MTF curves do not give a better or more precise answer than that in this context.

Jack

[Hans Kruse]

Great review which basically shows that the new Canon zoom lenses are up to the resolution of the 5DsR.

The final statement in the conclusion does lead up to another comparison: Finally, is it a medium format killer? No, medium format digital is
a different ball game both in terms of price, usability and performance.
It has some of the advantages of MF but keeps all of the
advantages of DSLR’s and the great range of Canon lenses.

[Josh-H]

Great review which basically shows that the new Canon zoom lenses are up to the resolution of the 5DsR.

The final statement in the conclusion does lead up to another comparison: Finally, is it a medium format killer? No, medium format digital is
a different ball game both in terms of price, usability and performance.
It has some of the advantages of MF but keeps all of the
advantages of DSLR’s and the great range of Canon lenses.

I found that last statement in the conclusion to be quite an underhanded swipe at MF. "Price" Canon Wins, "Usability" Canon Wins. Performance "?"

But the biggest take away IMO was the performance of the new range of Canon lenses. Superb.

[Hans Kruse]

I found that last statement in the conclusion to be quite an underhanded swipe at MF. "Price" Canon Wins, "Usability" Canon Wins. Performance "?"

But the biggest take away IMO was the performance of the new range of Canon lenses. Superb.

Yes, you may be right about the meaning of the MF comment. I would agree on that part very much myself.

Having upgraded to the newest Canon lenses 16-35 f/4L IS, 24-70 f/2.8L II, 70-200 f/2.8L IS II and 100-400 f/4.5-5.6L IS II I have been awaiting great performance on the 5DsR. I'm still waiting for my 5DsR due to lots of preorders and I waited too long to order mine. So the article was a great confirmation of that.

[bjanes]

My main beef with the article is using MTF curves in this latter context: they give the impression of a level of precision that is in fact not there.  One can pretty well make them come out as one pleases by playing with USM, local contrast and deconvolution sliders during conversion or in post (e.g. LR/ACR).  So if one can make them come out as one pleases, what's their use in a hardware comparison?  To make the results seem more 'scientific'?  They are no better than simply looking at the compromises off the pretty pictures themselves and deciding which appears to be more pleasingly sharp.  MTF curves do not give a better or more precise answer than that in this context.

Tim's Imatest results often show resolution above the Nyquist limit, and I find this problematic. The Imatest documentation states: 

"Sensor response above the Nyquist frequency is garbage. It can cause aliasing,visible as Moire patterns of low spatial frequency. In Bayer sensors (all sensors except Foveon) Moire patterns appear as color fringes. Moire in Foveon sensors is far less bothersome because it’s monochrome and because the effective Nyquist frequency of the Red and Blue channels is lower than for Bayer sensors. (This is found near the end of the post under the heading Some observations on sharpnes).

The slanted edge algorithm for calculating MTF is sensitive to sharpening and one can obtain impressive MTFs well above Nyquist, but when one looks at the images he may see sharpening haloes and Moire rather than improved image quality.

Regards,

Bill

[BAB]

 I rather like the thought of concentrating on what depth of focus can be achieved in the image from front to back in from edge to edge with this new Canon camera and the Canon lenses that are available for It.  It would seem that a 50 mb camera would be better used to make a large printed image, that being said take a 50 MB large sensor against the 50 MB small sensor make a print that is 30" x 40" comparing the two prints which one has more punch and less artifacts, which print has better sharpness side to side front to back and in the middle.  I would also think that one would have to gain access to several copies of the camera body and several copies of glass that was going to put on the front of his cameras to make sure that everything is in dropdead alignment.  I'm sure in the end if you could go through enough equipment to get your set the comparison be would be like splitting hairs but just to take a body off-the-shelf and any length and put it on it I hardly believe that that Canon camera is going to give you the same quality as you'll get from a medium format back.

[DeanChriss]

... I hardly believe that that Canon camera is going to give you the same quality as you'll get from a medium format back.

I'm sure you are correct, but the fact people are even making the comparison says something given that the camera and lenses cost a fraction of their MF equivalents. As with many things, at some point there's an enormous cost for a relatively small incremental step in quality. Whether it's worthwhile is always an individual decision.

[ErikKaffehr]

Hi,

With sharpening we get some egocentric MTF. This would essentially be OK and perhaps even relevant if there was a standard sharpening. So, it may be best to use like DCraw to decode the raw files.

I recall that Jim Kasson was doing some testing and found that the Sony he was using produced better MTF at large apertures than the Nikon. Stopped down they were pretty close. What Jim found out that although sharpening set to zero, LR was still applying significant sharpening, and that differed between the Nikon and the Sony. On the Nikon EXIF detected the lens in use. The adapter on the Sony gave no information on the lens. So, on Nikon LR had little sharpening as it knew that the Otus was very sharp, with the dumb adapter on the Sony LS assumed a mediocre lens and applied more sharpening.

Talking to all you gentlemen  gives great insights in weaknesses of testings and turn opinions into humble opinions. IMHO, that is.

Best regards
Erik

Tim's Imatest results often show resolution above the Nyquist limit, and I find this problematic. The Imatest documentation states: 

"Sensor response above the Nyquist frequency is garbage. It can cause aliasing,visible as Moire patterns of low spatial frequency. In Bayer sensors (all sensors except Foveon) Moire patterns appear as color fringes. Moire in Foveon sensors is far less bothersome because it’s monochrome and because the effective Nyquist frequency of the Red and Blue channels is lower than for Bayer sensors. (This is found near the end of the post under the heading Some observations on sharpnes).

The slanted edge algorithm for calculating MTF is sensitive to sharpening and one can obtain impressive MTFs well above Nyquist, but when one looks at the images he may see sharpening haloes and Moire rather than improved image quality.

Regards,

Bill

[Hans Kruse]

I rather like the thought of concentrating on what depth of focus can be achieved in the image from front to back in from edge to edge with this new Canon camera and the Canon lenses that are available for It.  It would seem that a 50 mb camera would be better used to make a large printed image, that being said take a 50 MB large sensor against the 50 MB small sensor make a print that is 30" x 40" comparing the two prints which one has more punch and less artifacts, which print has better sharpness side to side front to back and in the middle.  I would also think that one would have to gain access to several copies of the camera body and several copies of glass that was going to put on the front of his cameras to make sure that everything is in dropdead alignment.  I'm sure in the end if you could go through enough equipment to get your set the comparison be would be like splitting hairs but just to take a body off-the-shelf and any length and put it on it I hardly believe that that Canon camera is going to give you the same quality as you'll get from a medium format back.

Maybe I haven't seen it, but I have not seen alignment errors on my cameras and lenses. When you say small 50MP sensor sensor please realise that the size difference in less between 35mm and most MF like Leica, 645z and the MF cameras having the new Sony MF sensor, than the difference between 35mm FF and APS-C. I would agree with you that there could be alignment problems but do you have data to backup your claim that this is a real problem? I think both Canon and Nikon have their manufacturing process done in a way with very small tolerances. I would not expect that a medium format camera like the 645z would be more precisely done.

I assume that we are now again at cross roads where 35mm FF is closing the resolution gap to MF and where new sensors are being developed for MF that will increase resolution. However I think this time is different since we are at a point of diminishing returns.

[timparkin]

The calculation of MTF can show values greater than nyquist because the software knows that the linear edge in the test is just that. In a real world the camera wouldn't know if the edge was linear and hence would be unable to produce these results.

The test actually show the value of contrast below nyquist at MTF5 or MTF9 and hence I'm not sure what the problem actually is? I also keep an eye out to see if the contrast/resolution chart dips and then rises towards nyquist (a sign that things are going wrong).

Finally, the rock surface in soft, directional light shows specular highlights and broad levels of texture at many contrast ranges so makes a great 'general' test 'target' for my purposes. Happy to try something or at least discuss why you think it might be better?

Tim

[timparkin]

Not sure what the issue with including sharpening is considering the different cameras (and diffraction) introduce different levels of blurring. The idea is to introduce enough sharpening to cancel out this sharpening without overshoot. In fact Imatest includes algorithms that report on just this and hence have 'hard baked' compensatory sharpening into the software (as an option).

In a 'real world' we all use varying amounts of sharpening so excluding sharpening is more 'unreal' than including it. Discuss... :-)

Tim