Roger Cicala: Canon v. Nikon lenses on optical bench (no camera)

[ChristopherMarkPerez]

That would be great, except MTF is _not_ a measure of resolution. It's a measure of how much contrast a lens will transfer from the original scene to the film/sensor/imaging system.

MTF can give hints as to how humans will see resolution, but only at certain resolution frequencies that are determined by image size and viewing distance.

Contrast can be easily modified in processing. So it's never just about the lens.

Fortunately the authors of the site bring up other optical effects, such as field curvature. Chromatic aberrations and coma also effect optical performance.

There's a lot to this subject that is easy to get lost in the wrong or unimportant details. One has to be careful when attempting to understand what's really going on.

Fortunately, very little of any of this matters as commercially available optics and imaging systems are more than "good enough" when making very large, very sharp, very pleasing prints.

http://www.lensrentals.com/blog/2014/09/just-the-lenses-canon-vs-nikon-zooms-at-70mm

Roger at Lens Rentals is comparing 24-70's and 70-200's from Canon and Nikon. Interesting because it removes the huge variable of the different camera bodies you usually have.

 

[Rudeofus]

After reading yet another hatefest aimed at DxOLabs just recently, I am a bit surprised about the criticism aimed at Roger Cicala and his latest test report. DxOLabs got slammed hard because they measured lens performance together with the camera, and go figure, Canon didn't look good. Now Roger measured lens performance alone, Canon again didn't look stellar, and people throw another tantrum.

The only type of test, however contrived, which would find common acceptance here, would be one that yields results saying "Canon is better, pictures shot with Canon gear are automatically better, and people using Canon are a smart, attractive bunch, unlike users of other equipment."

 

[privatebydesign]

I'd like to know the actual resolution of a lens regardless of body.

Hearty Amen.

Why? Academia? Don't get me wrong, there is nothing wrong in and of itself that I can see, but seeing as how we pay thousands of dollars for these lenses that we can't use without bodies I question any results relevance.

Convolution.

Ask yourself this question. Why, when DPReview tests a body, do they use a quality prime at an optimal aperture, mounted to a heavy studio tripod, with careful focus bracketing and remote release?

The answer is, so that the lens so dramatically out-resolves the body that the results you get are almost entirely limited by the body itself.

Why do you want that? So that you can estimate how the body will perform with other lenses.

Same thing with lenses. How does the lens perform by itself? You want to know that so you can estimate how it will perform with any body.

If you don't do that, you're left testing every possible lens/body combination and retesting every lens every time a new body is released.

Convolution allows you to avoid that.

1/R^2 = 1/Rs^2 + 1/Rl^2, where R is system resolution, Rs is sensor resolution, and Rl is lens resolution.

If you know Rs and Rl independently, you can find R.

Convolution does not allow for lens variation, it can only give you a figure for the lens or lenses tested.

Convolution does not allow for factors most commonly used in actual photography, like auto focus.

Convolution is of limited value in telling you what you will actually get image wise.

Photography, despite the best efforts of many here, is still about images, not equations. Nothing will truthfully tell you what you will get out of your camera until you take an image with your camera and your lens in your shooting situation.

 

[privatebydesign]

And when we learn to take pictures without camera bodies the results might be relevant.

Yet more critical over analysis of a non relevant point. How a D810 and Nikon 24-70 f2.8 performs compared to a 5D MkIII and 24-70 f2.8 is all I, as an educated camera system buyer, want to know.

Science require control of the variables. Too many variables and you can conclude nothing.

Photography isn't a science. If you don't include all the variables you end up with purely academic test results that have extremely limited, if any, real world value.

In my opinion, the picture on the page is photography, but the benchmark performance of different optics is optical lens science. http://www.answers.com/topic/lens-1 There is no objective way to measure photography. But, there is, (and needs to be) a way to objectively measure lens performance. And, like all science, the best testing measurements require the best scientific method, which includes limiting variables to (hopefully)- one. Pax.

Why?

The logical conclusion from that train of thought is that the "best" tested lens will give the "best" photograph, when that is patently false. Even if you totally discount the photographer from the photograph making equation your fundamental point is flawed.

For example, you have an assignment to photograph Usain Bolt crossing the line of the 100m at the next Olympics, you can use any camera system, just get the image. For arguments sake the Canon 400 f2.8 tests better than the Nikon 400 2.8, but the Nikon has more MP and more DR, however the Canon has better AF, but the Nikon system can resolve more even though it doesn't test as well. Hm, the Canon can do more fps. Now which do you choose? The lens that esoterically "tests better" or the the other system that scores much better for more meaningful metrics; or the system that "wins" your test but for reasons other than that...........

I'm sorry, sir, but you simply cannot conduct science in the way you have described here. Once again, there are way too many variables. The fact is, that if you photographed Usain Bolt crossing the finish line with two identical situations except, one, with a good lens, and, the other with a great lens, the photo with the better lens will be better. That is simply a fact. About that there can be no rationale doubt or conflict.

Were I a Knight you could call me sir, as I am not you can't.

Were photography a science you might have a point, as it isn't you don't.

If you assume the AF, the AA filter, the demosaicing algorithm, etc etc have no impact on the image output your example might be valid, as they do, it is not.

 

[privatebydesign]

I'd like to know the actual resolution of a lens regardless of body.

Hearty Amen.

Why? Academia? Don't get me wrong, there is nothing wrong in and of itself that I can see, but seeing as how we pay thousands of dollars for these lenses that we can't use without bodies I question any results relevance.

Convolution.

Ask yourself this question. Why, when DPReview tests a body, do they use a quality prime at an optimal aperture, mounted to a heavy studio tripod, with careful focus bracketing and remote release?

The answer is, so that the lens so dramatically out-resolves the body that the results you get are almost entirely limited by the body itself.

Why do you want that? So that you can estimate how the body will perform with other lenses.

Same thing with lenses. How does the lens perform by itself? You want to know that so you can estimate how it will perform with any body.

If you don't do that, you're left testing every possible lens/body combination and retesting every lens every time a new body is released.

Convolution allows you to avoid that.

1/R^2 = 1/Rs^2 + 1/Rl^2, where R is system resolution, Rs is sensor resolution, and Rl is lens resolution.

If you know Rs and Rl independently, you can find R.

Precisely.

Precisely virtually zero practical value.

 

[privatebydesign]

After reading yet another hatefest aimed at DxOLabs just recently, I am a bit surprised about the criticism aimed at Roger Cicala and his latest test report. DxOLabs got slammed hard because they measured lens performance together with the camera, and go figure, Canon didn't look good. Now Roger measured lens performance alone, Canon again didn't look stellar, and people throw another tantrum.

The only type of test, however contrived, which would find common acceptance here, would be one that yields results saying "Canon is better, pictures shot with Canon gear are automatically better, and people using Canon are a smart, attractive bunch, unlike users of other equipment."

I hope you are not including me as a criticizer of Roger, several times I have written of my respect for him, his work, his results and conclusions.

I don't care what gear anybody uses, I use my selection because I felt it was the right thing for me to get, and I might point out that Roger is a 6D Canon system owner. I don't care how my gear "tests" I care how it works.

If your comment is directed in part at me I would also ask what is wrong in questioning the value in such a test, especially given that the tester himself has replied that he agrees it is of extremely limited, nay "pointless", value.

 

[Lee Jay]

I'd like to know the actual resolution of a lens regardless of body.

Hearty Amen.

Why? Academia? Don't get me wrong, there is nothing wrong in and of itself that I can see, but seeing as how we pay thousands of dollars for these lenses that we can't use without bodies I question any results relevance.

Convolution.

Ask yourself this question. Why, when DPReview tests a body, do they use a quality prime at an optimal aperture, mounted to a heavy studio tripod, with careful focus bracketing and remote release?

The answer is, so that the lens so dramatically out-resolves the body that the results you get are almost entirely limited by the body itself.

Why do you want that? So that you can estimate how the body will perform with other lenses.

Same thing with lenses. How does the lens perform by itself? You want to know that so you can estimate how it will perform with any body.

If you don't do that, you're left testing every possible lens/body combination and retesting every lens every time a new body is released.

Convolution allows you to avoid that.

1/R^2 = 1/Rs^2 + 1/Rl^2, where R is system resolution, Rs is sensor resolution, and Rl is lens resolution.

If you know Rs and Rl independently, you can find R.

Precisely.

Precisely virtually zero practical value.

Just because you don't understand it doesn't mean it's of no value to people who do.

 

[Rudeofus]

I hope you are not including me as a criticizer of Roger, several times I have written of my respect for him, his work, his results and conclusions.

I most definitely don't try to act as Roger Cicala Defense Force here, first off all I had my own stern exchanges with Roger in the past, second I can confirm from personal experience that Roger doesn't need anyone as RCDF because he is quite capable of standing up for himself.

I don't care what gear anybody uses, I use my selection because I felt it was the right thing for me to get, and I might point out that Roger is a 6D Canon system owner. I don't care how my gear "tests" I care how it works.

In this case the DxOLab tests are for you, and luckily CR will continue to link to them as they appear. Roger's tests can be used as complementary tests for all those, who fit Nikon lenses to their Canon cameras, or who fit Canon lenses to their Sony A7R, or as some have pointed out, for those who upgrade their cameras more often than their lenses.

 

[Michael Clark]

What it really means is there is the performance that can be wrung out of a lens if you really, really know your lens. Know where in the image it works best and where it works worst and plan your photographs accordingly. And that, my friends, is one of the things that separates great photographers from good ones. - Roger Cicala February 2020