LensRentals.com

"D" resolution tests

Posted by

It's been kind of an exciting couple of weeks, with 3 major new SLRs released and a couple of more on the way. There are plenty of people out there who are doing in-depth reviews, touting the greatness of the new cameras, and trashing them online without having touched one (My favorite so far is the guy who, after looking at online jpgs, stated it was obvious that the 5D Mk III and D800 weren't a bit better than his T2i, so he wouldn't be upgrading). I don't have much to say regarding image noise, ergonomics, autofocus accuracy, image processing, etc. There are plenty of people doing that more thoroughly and accurately than I could.

But there was one question that was really eating my lunch and I was in a position to take a look at it: just how much better would the Nikon D800, with that gazillion megapixels, really resolve? Would it be 3 times better than a D700, and 50% better than a Canon 5D Mk III, which the pixel count would suggest? Would the lenses we have really be able to take advantage of that resolution? I wasn't sure.

So when we got a bunch of Canon 5D Mk III's and a few Nikon D800s in last week and I was able to divert a few over to our Imatest lab for a few hours. There wasn't enough time to do exhaustive testing (generally the cameras arrived at 10 a.m. and had to be in packing to ship out by 3 p.m.) but I was able to get enough done to make some preliminary observations.

Comparing Camera Resolution

I arbitrarily chose two lenses to do the camera comparisons: the Zeiss 100 f/2 Makro Planar and Zeiss 25mm f/2.0. I chose Zeiss lenses because it let us put identical lenses in front of both Canon and Nikon cameras. These two particular lenses because both are exceptionally high resolution lenses and I wanted to be able to test at two different focusing distances, since that could make some difference. The copies used for this test had previously been tested and were known to be excellent and free of optical issues.

We tested each on D800, 5D Mk II, and 5D Mk III cameras (and one run on a D700 just for comparison). Otherwise things were kept as equal as we could make them: lighting and setups weren't changed, etc. Time constraints prevented doing what I would have loved to do: testing a half-dozen copies of each lens on a half dozen copies of each body. But this should be fairly accurate.

I should note that we initially ran the Canon files through DPP to convert the raw images, since Imatest can't directly convert the 5D Mk III files yet, but the results we got showed DPP was obviously doing some manipulation to the files as it converted them, making the results invalid for comparison since we test on unsharpened raw images. We then used Adobe's DNG converter which handled the files with no problems and didn't manipulate them at all, so we used RAW-to-DNG conversions for all the cameras to make sure things were equal.

The Zeiss 100 results first. The Vertical axis is the peak (center) MTF 50 (in line pairs / image height), the Horizontal axis aperture, and the cameras identified in the legend. The D700 and 5D Mk II results agree exactly with what we've seen testing these combinations for several months.

Results for Zeiss 100mm f/2. Makro Planar

Results with the Zeiss 25mm f/2.0 lens were very similar. I left off the D700 after the first test. I saw no sense beating a dead horse and, as I mentioned, time was short.

Results for Zeiss 25mm

The results certainly weren't surprising: I expected the 5D III to be a bit better than the II and it was. I expected the D800 to be better than any 35mm camera we'd tested, and it was. Previous, only the Leica M9, with its no-AA-filter, CCD-sensor, using the $6,000 Leica 50mm f/1.4 Summilux lens and gets up in this range among 35mm cameras. For the couple of people, though, that seem to think the D800 is a medium format camera in 35mm clothing, I would point out that a Hasselblad H4D-50 with kit lens tests out at about 1,600 lp/ih, so no, we're not quite there yet.

At Higher ISO

The above results are taken at ISO 200 which should theoretically giving best, or near-best, performance for each camera. I was curious how the D800's resolution would hold up at higher ISOs so I repeated the ZF 25mm on D800 series at ISO 400 (where I do most of my shooting - it's my test after all) and also at ISO 1600 and ISO 3200.

Again, this is done from raw images with no (as best I can determine) post image sharpening, although you can never be sure what is happening in-camera. But at any rate, there really is an amazingly small amount of resolution fall off at reasonably high ISOs. I was really surprised at this, especially at how well 3200 compared to 1600. Obviously I should have gone further, and need to do the same comparisons for the 5DIII, which I should get to next week.

What About Lenses?

Ah, now that is the question. At least it's the question now. Lloyd Chambers had already mad some good suggestions for Zeiss and Nikon lenses that should be able to handle the D800's resolution based on his experience. I'm not sure I agree with all of them, but it's certainly the best starting list. I plan on testing each lens on the D800 and getting a list of our own together, but I was able to get some of the usual players tested before the last D800 left the shop.

The first graph plots peak (center) MTF 50 comparing the ZF 25, Nikon 24 f/1.4G, Nikon 14-24 f/2.8, and Nikon 70-200 f2.8 VR II. There's a pretty significant difference between the primes and zooms at f/2.8. It's not surprising, since the zooms are wide open there, but I thought the point was worth making: if you want best resolution with the D800, shooting at f/4 or f/5.6 is going to be necessary with most lenses.

Center only MTF 50

The second plots average resolution of the center, halfway to the corners, and corner MTF 50.  It becomes apparent that center resolution doesn't mean corner resolution: the 70-200 VR II does much better in the corners than the 24 f/1.4. The Zeiss 25mm does superbly well, but I should point out that this lens seems to do it's best work at close and medium distances (like where it is when we do Imatesting) and may not be as good at infinity.

Obviously, there are a lot more lenses that we'll need to test just to make recommendations based on resolution. The only message I think to take away right now is that the D800 is playing up in the range of maximum resolution of even the best lenses. Putting anything less in front of it is going to limit the camera.

Conclusion:

For the fanboys who don't like the results: This concludes our test of the Emergency Resolution Testing Service. This was only a test.  If this had been an actual Fanboy emergency you would have been instructed where to tune in your area for official Fanboy disinformation and complete manufacturer sponsored reviews.

For everyone else, there's no question the D800 can actually get those pixels to show up in the final product (assuming your final product is a big print - they're going to be wasted posting on your Facebook page). But you'd better have some really good glass in front of it if you want to demonstrate all of that resolution.

In the real world, highest possible resolution is nice to know about and talk about, but usually not of critical importance compared to other factors. You'll be able to make superb images with any decent lens for an 8 X 10 or even 11 X 16 print. But if you're getting the camera because of the resolution, it makes sense to know which lenses will allow all of that resolution to be utilized. Just in case you get that job that needs billboard sized prints.

 

Roger Cicala

Lensrentals.com

March, 2012

75 Responses to “"D" resolution tests”

Daniel Browning said:

* Jerry Russel wrote:
> How do you do your sharpening?

Usually just Lightroom.

> Do you have an application that does “diffraction deconvolution”?

I don't shoot in diffraction-limited circumstances very often (I'm more of an f/2.8 kind of guy), but I'm still looking into it. My front runner is Raw Therapee, which uses the Richardson Lucy algorithm.

That said, I may have used a similar deconvolution in this case. IIRC, Eric Chan (one of the software engineers behind ACR/Lightroom) said that the sharpening slider moves between USM and deconvolution sharpening with a gaussian PSF. I'm not sure how much different that is from a diffraction PSF (probably just models the central spot and no rings), but I vaguely recall someone saying they were similar enough to be used for the same purpose. Somewhere in the middle (which is where my setting was for this shot), the slider uses a mix of both.

> Your sample looks great.

Thanks. For that I one I used 45/1.0/100 in LR4.

> I think I see the “moire” — you’re referring to the broad, dark banding, right?

Yes.

> Wouldn’t you expect to see at least some of that type of artifact, unless you filter hard at half the nyquist rate?

Yes, but my feeling is that I'm seeing it more and easier with the D800 than the 5D2.

> I’m not sure I understand why you would expect the results of this test to be different at MTF30 or MTF15.

Because my understanding is that most lenses have long-tail MTF curves. Their MTF drops pretty precipitously for the first 30 lp/mm or so, but then it holds on to the last bit of MTF for a long time. But perhaps I'm mistaken.

> DXOmark tests lenses at MTF20, and if you look at their results for the Canon 85 1.8 (sharpest Canon lens they’ve tested) they get 67 lp/mm on the 1ds3 (pixel pitch 6.4 um) and 80 lp/mm on the 7D (pixel pitch 4.3 um). That is, a 48% increase in linear pixel density on the 7D (2x the data per square millimeter of sensor area) buys a 20% improvement in resolution at MTF20.

I must be mistaken, then. I thought that hitting MTF-15 at frequencies nearing the Nyquist of a 4.3 micron pixel wasn't really that exceptional, but perhaps it is. I'm glad that my $90 macro is one of them. It's hard for me to find the data you used on dxomark, their site has been very slow lately.

Can you check their data for the 300mm f/2.8 II? That thing is almost diffraction limited, and even with a 2X TC turns in nearly flawless resolution. But if DxOMark thinks it's no better than the 85mm f/1.8, I think there must be something else going on, because I'm pretty certain the 300mm can give the full expected return of 4.3 micron pixels.

> This also highlights the fact that it’s just as important to use high resolution lenses to get the most out of any recent crop camera, as it is to use great lenses on a D800.

Agreed.

Rob said:

Brian Potter - the 24pc-e is sadly not fully compatible with the D800! the popup flash gets in the way of movements.

Also it looks like the live view 100% zoom is acting a bit weird too, which isnt great for tilt shift use...

Duncan D said:

Thanks for the tests can't wait to see which lens are fit for this big boy

Frans van den Bergh said:

*Daniel Browning wrote:

> I’m not sure how much different that is from a diffraction PSF (probably just models the central spot and no rings), but I vaguely recall someone saying they were similar enough to be used for the same purpose.

This depends on what aperture you are referring to. At larger apertures (say, f/2.8 to f/5) the MTF curve of diffraction looks very much like a straight line (see my earlier post for a link with examples). In other words, it decays much more slowly with increasing frequency, compared to the Gaussian approximation of the sensor AA filter.

The RL deconvolution implementation in RawTherapee is hard-coded to use assume a Gaussian PSF. This implies that for a small radius (standard deviation) setting of about 0.57 in RT's RL sharpening, you can "undo" the Gaussian blur of the AA filter, but diffraction will remain. If you use a larger radius, you will sharpen more, effectively reducing the effects of diffraction, but also introducing significant distortion below Nyquist.

Daniel Browning said:

Ah, thank you for the explanation!

Jerry Russell said:

Daniel: Thank you! DXO says the 300 2.8 IS II gets 61 lp/mm on the 7D. Looking through more of the DXO lens data, I agree there's something odd going on. They think the 70-300L is not any sharper than the old 70-300 IS, for example.

George Barr said:

Information much appreciated. Wonder where the Pentax 645D and say, the 120 macro would fit in the graphs.

Bryan Willman said:

Do you have the same data for an M9, and could you overlay it with the D800?

As an M9 user, and geek, I've always been a little perplexed by the high ratio of "picture success" to "technical specs" - and just think "it's the lenses" and shrug.

Also - you test suggested that for the D800 and the 5dM3, resolution is lens limited wide open, but becomes sensor limited by about f/4 for these particular lenses. I wonder if we're coming to an age when 1st line cameras will be limited lenses rather than sensors. (And both will be swamped by technique.)

Dave Sucsy said:

Hey Roger,
Thanks for your great business and your great blogs and testing info. Also your sense of humor. We might as well have fun while doing our work!
I'm annoyed with both Nikon and Canon. The D800's versatility of the file size (croppability, down-res ability, and better dark value manipulability) may cause me big trouble in having to switch from Canon to Nikon. And Nikon is really holding my nose to the grindstone in causing me to have to be more picky than ever with my technique. All I do is real shooting for a living.
Thanks again for this objective performance report.

Pavel Vaisberg said:

Hey!
Thanks for the testing.
Trying to figure out is the D800 will give any advantages in macro (real one :) ) where effective working aperture is 30 and smaller. Is any tests showing the graph until f22 (by settings)?

Thanks again!

Daniel Browning said:

Pavel, it depends on which camera you are upgrading from. If it's the D700, then there most definitely will be a significant detail increase possible with the D800, even when both are compared at f/30. If it's from the D3X, which is very near to the diffraction cutoff frequency for f/30 (i.e. MTF = 0% for green light), then the difference in detail will be nothing (the only possible improvement will be due to contrast variation between the two OLPF).

malchon kao said:

What is your comments by using old full frame lens as Hasselblad connected to D800?
I collected 4 lens but really like to listen to you.
I did not use very often since overlap the range of Leica R lens.

Thanks

LensRentals Employee

Roger Cicala said:

Daniel,

Feel free to post anything you like here. I want it to be a bit of a refuge for thoughtful commentary like yours without fanboi drivel.

And let me suggest that everyone follow the link to Daniel's extremely well thought out, factual, and impartial review. I could not have done it as well but I agree with it completely. It captures the 'essence' of the camera very thoroughly.

Now I've got to get Daniel to start writing some blog posts over here :-)

Roger

Daniel Browning said:

Thank you, Roger! You're too kind.

John D said:

I know this comment comes way late in the game but was wondering what the results would be on a non Ivy-League lens like an 18-200. The reason I ask is because I am always asking myself "compared to what?" For the tests you did above, I think the comparison is (keeping my metaphor intact) "Comparing the Harvards to the Yales."

It got me wondering what a state school could do.

LensRentals Employee

Roger Cicala said:

John,

Like most lenses, it will give more resolution on a D800 than on another camera. But the difference between it and another lens may (may) be magnified.

Simone said:

Roger,

wanted to say thanks for this interesting article.
Impressive performance from D800, more so if one considers how much the effect of sharpening is boosted by increase in resolution.
I tried to quantify it with a simulation scenario about sensor performance that I completed before actual full frame DSLRs came out. (full frame 20mpxls, 36mpxls and 50mpxls). Empirical data you posted seem to confirm it.
I think resolution is not one of the main factors involved in producing and printing, good pictures; but most of us crave for performance when buying a new top-end body, so knowing how much boost we can expect can offer a guideline for upgrade.
Im posting the link in case you may be interested in knowing how far a 50mpxls body would get; the full analysis is in the attached pdf.
The development is a bit math-oriented but final results are readily usable.

http://www.canonrumors.com/forum/index.php?topic=2412.msg51441#msg51441

Regards

Simone

Terence said:

Great work and interesting results. But can you explain why your MTF 50 figures are about one third of Photozone - and I think they used a Nikon D3 to test the 100 mm Zeiss. This is just a technical question not a criticism. I thought the MTF 50 in lp/h was a standard used to enable independent comparisons. I would expect slight variations in measurement, but not 3x. I've pasted the link to the Photozone test.

Thanks

http://www.photozone.de/nikon_ff/590-zeisszf10020ff?start=1

LensRentals Employee

Roger Cicala said:

Terence,

Photozone is using Line widths per image height, I believe, I use Line pairs / mm. So you would double our line pairs / image height results to get line widths. (A black or a white line is one line width, and black and white line is one line pair.)

They also test jpgs, while we're testing raw images. jpgs are always going to be sharper (we see about 50%). One's not better than the other: Photozone is showing what the image should resolve out-of-the-camera. We're trying to eliminate as many variables as possible for testing purposes.

Of course, there's still going to be some variation: which charts are used, what distance was tested, copy-to-copy variations, etc.

Roger

Terence said:

Thanks for that clarification, Roger - it makes perfect sense.

Regards,
Terence

Carles Mitja said:

Dear Roger, I have a couple of questions.

1) What kind of units are those in the vertical axis of published plots? I understand MTF50 but I'm not sure about the vertical axis units.

2) When you said "We then used Adobe’s DNG converter which handled the files with no problems and didn’t manipulate them at all, so we used RAW-to-DNG conversions for all the cameras to make sure things were equal.", What it really means? Did you deactivated or zeroed all ACR tabs? o did you preserved the default adjustments?

Thanks a lot for your time.

Carles

LensRentals Employee

Roger Cicala said:

Carles,

All axis are measure in line pairs / image height.

We zero all adjustments.

Roger

Carles Mitja said:

OK, Roger I understand. I will assume that you have measured MTF for "horizontal" picture height.

Then, your plots are comparing things that I think cannot be compared. A result of about 1200lp/PH for the MTF50 value in the case of the D800 is WORSE than a result of about 1050lp/PH for the same MTF50 in the 5DMKII case. It's been comparing absolute units from each camera in a relative fashion. It is more or less obvious that the D800 with its higher absolute resolution tends to proportionate a higher ABSOLUTE result.

I have reduced the results for the MTF50 from your plots to relative units, using it as a percentage of the resolution capabilities of each camera, and the results are:

- D800. MTF50 at about 1200lp/PH, it means MTF50 at 32.6% of its theoretical capability.
- 5DMKIII. MTF50 at about 1050lp/PH, it means MTF50 at 36.5% of its theoretical capability.
- 5DMKII. MTF50 at about 1010lp/PH, it means MTF50 at 36.% of its theoretical capability.

The higher the frequency where the MTF50 is achieved, the best the image quality performance. It is true that the final differences are not so important in terms of pictorial results and its influence will depends on the scene contents in terms of frequency or details, but they are quite different than those explained in your post.

After all, the D800 lost in contrast is a fine result for a camera with such small photo-receivers.

Do you agree?

Carles Mitja said:

P. D.

I have selected the f/4 Zeiss 25mm results to do my calculations because is the best one achieved for the three cameras.

Leave a Reply