Sorry to ask: Highest "resolution" lens for D800?

FlowtographyBerlin

Ok, but since some of you have already pointed out the issues with the 105 VR in non-macro distances, that's not a good comparison. I'd think any zoom will be beat by an excellent prime within its focal lengh range. Which is only logical.

FlowtographyBerlin

The benefit of the lensrental data is that all of those MTF numbers were obtained using the same setup & methodology on the same camera (D800 in this case).

Still (and then, especially), as I wrote above, the numbers don't make sense to me because they imply the sensor outresolves the lens. Or what am I getting wrong?

JJ_SO

don't know how thin a line (measured in pixels) may be to be recognized as a line. But logically, 1/3 of the theoretical resolution doesn't make much of sense as there are no "half pixels" to build lines with.

I think 1/3 resolution means than you have one pixel for each colour? That would be 1637 1/3 RGB pixel groups (but the Bayern pattern doesn't equal RGB, there are more green dots, so the reproduction of a black line actually needs 4 pixels in one direction). And to make a difference from line to no line (=LinePAIR) it takes theoretically another 4 pixels = 8 pixels. Now 4912÷8 = 614 linepairs. Guess, there must be something wrong with my assumption, so I wait for people who really know instead of going on wit my guessing. In each case the question remains how to get an exact number of linepairs?

What I learnt as well in Roger's excellent blog: There's only a clearly visible difference if there's something like 20% less LP/IH. Or, if I may quote him:

"One thing I suggest you do right now: downsize the image to 50% instead of 100% (which most people suggest as a more reasonable approximation for viewing an actual print). You can probably still see the difference. Now downsize it to 12.5% (about online jpg size). Can you see the difference now? If you can, you need a medication increase"

Ade

1. Here are some pairing data to consider when discussing primes vs. zooms (from the same database):

Prime: 24/1.4G resolution center: 1183 average: 845
Zoom: 14-24/2.8 resolution center: 1181 average: 831

Prime: 50/1.4G resolution center: 1073 average: 889
Zoom: 24-70/2.8 resolution center: 1107 average: 902

Prime: 85/1.4G resolution center: 1044 average: 1021
Zoom: 70-200/2.8 resolution (normal distance) center: 1080 average: 957

As you can see, the pairs are within statistical variance. I think the myth that primes beat zooms is busted (in terms of resolution, at least).

2. There is no implication that the "sensor outresolves the lens"

First, remember that these are MTF50 numbers used for relative comparisons. They are not absolute resolutions. Much higher resolution numbers could be obtained at different (lower) contrast thresholds and for different setups. MTF50 is generally selected because it corresponds well to our perception of sharpness.

Second, the MTF test measures the combination of the camera + lens together. It is not a situation where one part is "limiting" or "out resolving" the other part. Maybe it's more correct to say that both parts are limiting the total system.

If you replace the camera with a better one, the MTF improves (so the camera isn't the limiting factor). But if you had instead replaced the lens with a better one, the MTF also improves (so the lens isn't the limiting factor, either).

I'm going to simplify quite a bit here, but to help understand this better, maybe think of MTF as an efficiency factor. Suppose you have a camera sensor at 80% efficiency, and a lens at 90% efficiency. The combination yields (0.8 * 0.9) = 0.72 (72%) efficiency.

Which one is the limiting factor? That's irrelevant. An improvement on either the lens or the sensor would improve the total efficiency.

Paperman

" The benefit of the lensrental data is that all of those MTF numbers were obtained using the same setup & methodology on the same camera (D800 in this case) "

Same must be valid for other test sites. Their results are also more comparable to each other.

Ade

Unfortunately, no. Look at the DP Review tests, for example. Some FX lenses were tested on a Nikon D3, while others were tested on a D3X.

Ade

I just checked the Photozone database. Most of the FX tests were done on a D3X. However, they did their 105 VR test a long time ago on a D200 (not comparable obviously).

They do have recent test results for the 60mm macro, 85/1.4, 24-70 and 70-200, all on the D3X:

Macro: 60/2.8 resolution center: 3825
Zoom: 24-70/2.8 resolution center: 3988 (at 24mm), 3854 (at 40mm), 3706 (at 70mm)

Prime: 85/1.4G resolution center: 4016
Zoom: 70-200VR2 resolution center: 4028 (at 70mm), 3929 (at 135mm), 3859 (at 200mm)

Again, we see:

- Macros are not necessarily sharper at non-macro distances (the 60mm's res was the lowest max out of the four lenses presented)

- Primes are not necessarily sharper than zooms (both zooms held their own vs. their prime counterparts)

So we now have the same conclusions from both the lensrental and Photozone databases.

KnockKnock

Just curious Ade, are those numbers on the primes shot at the equivalent f/stop (f/2.8 presumably)? Or are they representative of f/1.4? Or just max numbers wherever the lens excelled?

spraynpray

It is hard to come to any conclusion reading the last part of this thread because there is so much missing information regarding test conditions - what distance are the tests done at? It does seem weird to me even trying to compare a micro with a non-micro if one is optimised close up and the other further away?

Also @Ade: you said "If you replace the camera with a better one, the MTF improves (so the camera isn't the limiting factor). But if you had instead replaced the lens with a better one, the MTF also improves (so the lens isn't the limiting factor, either)."

Seems to me that is just wrong because if changing the camera for a better one improves the result, it DOES mean that the camera is limiting the result.

Am I missing something (Wouldn't be the first time) ?

Golf007sd

Having read all the a above responses, one those have to look at this topic with some reservation. Does lens to body calibration play a roll as well? Or is it that the sensor within the D800 the key variable in relation to the lens?

Ade

@KnockKnock:

Typically the lens performance is tested at various f/stops, but the final number reported is at whatever f/stop the maximum resolution was achieved.

Photozone actually shows the individual f/stop results. The 60mm Macro's max center resolution was at f/5.6, while the 85/1.4G's max resolution was at f/4.0.

@spraynpray:

1. The test distance, etc., are largely irrelevant -- as long as they are consistent from test to test.

The test distance specifically is not critical anyway, providing the test chart adequately fills the frame (more or less, above a certain minimum). Generally, several different distances are used depending on the focal length of the lens. E.g., the wide-angles might be tested at one particular distance, while the mid-range lenses are tested at a different distance.

2. The macro comparison is in response to TTJ's remark that his 60mm macro is "almost too much" (resolution) for portrait work. While that might indeed be true, as we've seen the resolution of the 60mm macro is actually less than many other lenses -- including the portrait-specialist 85/1.4 -- at non-macro distances.

3. The camera in that example would indeed be "a" limiting factor, but it isn't "the" limiting factor (i.e., the bottleneck). Sorry for the nuance there.

If the camera is truly the limiting factor, then a better lens would not improve the result. The only way to improve the result would be to replace the camera with a better one.

But that's now how real life MTF works. A better lens would indeed improve the overall MTF. The camera would only be the limiting factor if we had the perfect lens (which we don't). And on the flip side, the lens is also not the limiting factor, because we don't have the perfect camera.

If you re-read what I wrote in the previous post about efficiency factors, then you can see how increasing either factor would increase the total result -- so no one factor is the limiting factor.

@Golf007sd:

Lens calibration is not usually relevant for lens testing because in most cases the lens is focused manually against a focus target. Live view is used to "zoom in" and critically check the focus.

Focus bracketing is then used on top of that. E.g., the tester will actually take several images at slightly different focus distances. The resulting MTF chart is derived from the image which yielded the highest resolution.

So AF performance doesn't play at all into the lens measurement results. I hope this clarifies the issue.

spraynpray

Is it really irrelevant Ade? If a lens is tested at a distance it isn't designed to be best at, it isn't a fair comparison. If the distance is different, it should be stated. My point was that is isn't stated so how do we know what's what?

I have a feeling that we are all doing the lens equivalent of pixel peeping here, but it is interesting! :-B

Ade

Well, practically speaking, a special test distance is only required for macros. But that is why two different measurements were given for the 105VR: one for macro distances, one for non-macro distances.

But it is still irrelevant anyway, because at both distances the 105VR's measured max resolution was still less than the other lenses. So the results didn't change with distance.

We can't test a chart at infinity (you'd need a very big chart) so in all other cases the test distance is selected so that the test chart fills the frame, but isn't so close that the quality of the chart itself becomes an issue.

Imatest isn't really sensitive to the exact test distances. As long as the test chart isn't too close, the software can compensate for differences in distance.

FlowtographyBerlin

Oh dear, this is starting to get really messed up... Lots of confusions, lots of missing information, lots of mis-information. Also on my side, no exception here. But just some things:

don't know how thin a line (measured in pixels) may be to be recognized as a line. But logically, 1/3 of the theoretical resolution doesn't make much of sense as there are no "half pixels" to build lines with.

The genereal "rule of thumb" for sensor resolution is Pixelcount roughly MINUS a third, so you have MORE than one pixel to build a line. Theoretically, you need one pixel for a line, of course, less is not possible physically. It is because of the Bayer array that you don't get full resolution for every pixel. And no, the calculation is not one third left over, but one third OFF. That's a rule I learned a little longer ago and it should be available on the more serious websites dealing with this.

If you have an image height of 3000 px, it would be 2000px, converted to line PAIRS, that makes it 1000 LP/PH.

@Ade: you said "If you replace the camera with a better one, the MTF improves (so the camera isn't the limiting factor). But if you had instead replaced the lens with a better one, the MTF also improves (so the lens isn't the limiting factor, either)."

Seems to me that is just wrong because if changing the camera for a better one improves the result, it DOES mean that the camera is limiting the result.

Exactly. The camera used definitely IS a limiting factor. Of course. Because it delivers more or less resolution. The 50 1.4D on a D700 was razor sharp in situations that looked blurry on my D800.

It is exactly because the camera now makes every little bit of flaw visible that people get into pixel peeping. Of course, it can get ridiculous. I'm with you. As @Msmoto said so nicely "most lenses outperform my abililities". True in real-life photography for almost anyone, I'd say, including myself, and including all the pixel peepers who think better equipment will get them better pictures.

On the other hand, if you have a situation where you want to take full advantage of the high resolution, this does become an issue. And for me, there sometimes ARE these situations, otherwise I'd buy a different camera. I love the high resolution of the D800.

Having read all the a above responses, one those have to look at this topic with some reservation.

Definitely, but more because there's so much confusion.

@Ade, thanks for all your explanations and efforts. Still, the values from lensrentals don't make any sense to me, because it indeed would imply what I wrote before, that the lenses don't even reach the resolution of the sensor. I'll try to find a source for that sensor resolution thing I wrote above and post it here.

Concerning the prime vs. zoom issue, I of course believe your research results, but I still would stick to the point that there will always be a prime outperforming all these zooms that you mentioned. Suggestions: Zeiss 15mm, 21mm, the Leica 28/2.8 (I think it was that one), the Sigma 35mm 1.4 (just an idea), one of the Zeiss 35mm, maybe the Nikon 35mm. The 50s are a bit difficult, so I'd believe that, but on the other hand I've seen images from the 50 1.8 compared to the 24-70 at 50mm, and you could see the difference (pro 50). Also, the 85/1.8 I can't believe it will be outperformed by the zoom. The Zeiss 85 and 100? And so on.

Disclaimer before you start ranting: I haven't used most of the above lenses, so it's merely a guess; I have seen the performance (not test chart numbers) of all those in image comparisons, though, which is what I'm basing my guesses on :-)

Ade

@FlowtographyBerlin

I've mentioned it before, but the MTF50 numbers do not indicate absolute resolution. In fact, the numbers themselves are meaningless except for comparison with other numbers derived with the exact same test methodology.

Think of the lensrental test as being 'harder' than the photozone test, due to methodology differences. E.g., different test charts, lighting conditions, RAW processing, whatever.

In this case, we would expect the same camera+lens combination to score lower in the lensrental test, vs. the photozone test. In equivalent units, maybe a camera+lens that would score 1200 in a lensrental test might score 2000 in the photozone test.

And you could setup your own Imatest rig, and using your unique methodology, find out that the same exact camera+lens might score 1600 on your test.

Which number is correct: 1200, 1600 or 2000? They're ALL correct!

I hope you can see that the resulting numbers are completely arbitrary!! They don't relate to absolute resolution. They are for making relative comparisons.

So we cannot take these MTF50 numbers and compare them to theoretical sensor limits and make claims like the sensor is out-resolving the lens.

TaoTeJared

The MTF debacle above needs to be moved to it's own "MTF Thread".

Golf007sd

Very healthy conversation. Thank you all for your input.

In the end, I'm pleased to say: I have almost all the lenses mentioned. Moreover, I hope each of us that own such amazing glasses are able to take the best shots possible give our skill level. The imporvment will be achieved by us all in mastering our technique in order to take full advantage of our gear. Let the chips fall where the may...the MTF chart can kiss me where the sun does not shine.

FlowtographyBerlin

I've mentioned it before, but the MTF50 numbers do not indicate absolute resolution. In fact, the numbers themselves are meaningless except for comparison with other numbers derived with the exact same test methodology.

I hope you can see that the resulting numbers are completely arbitrary!! They don't relate to absolute resolution. They are for making relative comparisons.

So we cannot take these MTF50 numbers and compare them to theoretical sensor limits and make claims like the sensor is out-resolving the lens.

@Ade: Thanks for taking the time to explain again. I wouldn't agree with you that the numbers are completely arbitrary, though; you're talking about the LP/IH values like they're something completely abstract, but I'd say they're not really. Yes, as you said before, the numbers depend on the combination of camera and lens, which is natural because the camera is the device that is actually recording the whole thing.

There are two things in those values: Line Pairs, and Image Height. A line pair will need at least two pixels. Image height is the height of the sensor. There is a resolution that the sensor itself delivers, measured in line pairs. If the measured line pairs that a lens resolves on a specific camera is lower than the sensor of the camera, this logically implies that the lens delivers less line pairs than the sensor could resolve.

Of course, the values are going to differ from one measurement method to the other, but after all, they do stand for a concrete value, which is line pairs per image height. Differences of 1627 vs. 1750 and the like: For these, what you said applies, because one method and rig will produce a different result. But 1000 vs. 1600 line pairs on the same camera, that's a whole world apart and again, the 1000 line pairs on a sensor that delivers 1600, that implies... blah blah blah, I wrote it all before :-)

In the end, I'm pleased to say: I have almost all the lenses mentioned. […] Let the chips fall where the may...the MTF chart can kiss me where the sun does not shine.

:-)

Michael_Erlewine

I am a close-up/macro shooter and have had most digital Nikons since the beginning. Right now I am using the D4, 800E, and 7100. I have some 80 lenses, most of them very sharp, the best not Nikon, but Leica, Voigtlander, and so on.

My point in writing here is to point out that in my experience the quest for sharpness turns into a question for APO (apochromatic) lenses. In other words, once you get really sharp lenses, then the difference becomes IMO how well they are corrected. If a lens is sharp and then has all kinds of fringing issues, the result is not what I call sharp. Sharp corrected lenses I know of include:

Voigtlander 125mm f/2.5 APO-Lanthar
Leica Macro 100mm APO Elmarit-R
Leica Macro 60mm Elmarit-R
Nikkor El-Pro APO 105mm
Printing Nikkors (75mm, 95mm, 105mm, and 150mm)

So correction is what determines the difference, all things being equal.


I am the close-up mentor at NikonGear.

Another factor is the presence or absence of the AA filter. My D4 has one, my D800e has almost none, and the D7100 has none. These cameras are sharp in reverse order, with D7100 being the sharpest, i.e. no filter at all.

SquamishPhoto

Nice post, Michael. :]

FlowtographyBerlin

Michael, thanks for your post! Interesting suggestions!

Msmoto

@Michael_Erlewine

Thank you so much and welcome to NRF.....of course now I will run off and investigate some new goodies...

Golf007sd

@Michael_Erlewine: From my perspective, "shapness", it is the lens that is the primery factor not the sensor on a given body. The sensor address how it can capture the image itself given the available light. Moreover, how well it controls noise, hence ISO.

SquamishPhoto

Clearly his rather esteemed background as photographer and author on photography and rather significant collection of lenses and bodies would lead one to think otherwise, Golf. Im not sure how you come to your conclusions, but it doesn't seem terribly scientific.

http://macrostop.com/

Golf007sd

@SquamishPhoto: I'm not question his knowledge on gear or style of photography. My point being in relation to the topic at hand. Hence, on a D800, it is the lens on the body that will yield the sharpest image vs the sensor of the D800 itself.

Michael comes from a perspective that he believe "sharpness" depends [comes] from color. In fact, he questions himself weather he is crazy for thinking so. I do not see sharpness in that light. The dynamic range of the D800 is fantastic and does present and amazing level of detail captured by the sensor...but without a quality pro-lens, this level of clarity would be lacking. Hence, my conclusion.