NEX-7 Rolling Review

[Mark D Segal]

Assuming that you shoot raw, you can just lower the contrast in the NEX-7/5N EVF and it opens up shadows quite a bit.  It gives you a more accurate histogram for the raw, too.

Douglas, could you please point me to how one does this. I have the NEX 5n and the EVF. The only instruction I could find is on page 143 of the 5n manual where they say the brightness of the finder can be varied by +/- 1, whatever that means, but nothing about contrast or the impact on the histogram.

[01af]

Michael's NEX-7 vs M9 comparison is badly screwed up, invalid, and thus, entirely pointless.

I do appreciate the effort, and I do agree that the latest generation of Micro-4/3rds and APS-C-format mirrorless cameras are a serious competition for the M8 and M9 cameras when used with M lenses. Actually, I believe Leica must, and will, develop some kind of "NEX-7 look-alike with 35-mm-format sensor and M mount" pretty soon.

However this partcular comparison is not worth the electrons required to transmit them. It would be better to take it down from this site because it's not only invalid but highly misleading. Michael's word has some weight in the world of today's photography, and he should not put his reputation at risk with such a nonsense.

Michael wrote, "I cropped the M9 images to the same field of view as the NEX. [...] I then resized the NEX files from 24 MP down to 18 MP so that they matched in both field of view and resolution."

After cropping the M9 files, to match not only the field-of-view but also the resolution, the NEX files must be downsized from 24 MP to 7.8 MP, not to 18 MP.

However, this is not the worst thing about this comparison yet. Even worse than this hard-to-believe arithmetical error is the fact that the M9 shots are out of focus. Hey! How can you dare to compare test shots when one set is properly focused and the other is not?

For the six shots (three with NEX-7 at f/1.4, f/2.8, and f/5.6, and three with M9 at the same set of apertures), the same lens was used from the same shooting distance. So any differences in sharpness at the same aperture can only come from the different sensors, right? However, in the NEX shots, the sharpness at the frame's center improves only marginally when stopping down; in the M9 shots the sharpness improves significantly when stopping down. How can the sensor affect the lens' response to being stopped down? Simple—it cannot. Instead, the M9 was slightly out of focus, and at f/2.8, increasing depth-of-field compensated partly; at f/5.6, almost fully.

Of course, it is to be expected that in NEX-7's higher resolution will lead to more detail in the final image, even after downsizing to the same pixel count. But the difference is by far not as great as the M9 comparison shot at f/1.4 suggests. Moreover, when Michael really was interested in image quality rather than pixel peeping then he would shoot his test with equivalent lenses (i. e. 35 mm on NEX and 50 mm on M9, for example) and then resize both full frames, without any cropping, to the same final size for presentation, be it digital files or prints. This way, you'd assess actual, real-world image quality rather than academic per-pixel quality (or here: per-sensor-area quality), and the M9's larger sensor area would be able to compensate, to a degree, for the lower pixel count. Throwing away the M9's biggest advantage over the NEX-7 is not a good start to arrive at a valid (from a photographer's point of view) comparison.

So if this screwed-up comparison teaches us anything then it's this: The opto-mechanical rangefinder is outdated by now. It cannot provide the precision that is required to reliably exploit the full potential of modern lenses on modern sensors. Accurate focusing with a manual-focus lens is much easier to achieve with the NEX-7 than with the M9. I believe the next Leica M will include some sort of electronic focus confirmation. The days of the traditional rangefinder are over.

[pservice]

I agree with other comments that, given the way the test was done, it would have been "fairer" to downsize the NEX-7 image to 7.7 MP, which is the approximate size of the M9 crop.  Still, I think it would be interesting to do the comparison when both sensors are recording the same field of view.  Which obviously requires changing the camera-subject distance or using different lenses.  Either way, I would expect the NEX-7 image to have higher resolution.  Because it would be sampling the scene at higher frequency.  Although it might depend on the amount and kind of detail present in the scene.

[Mark D Segal]

Is it possible maybe to do the same comparison versus NEX-5n maybe. I would be very interested in this so that we (owners of NEX-5n) can decide if it makes sense that at some point in the future we switch to NEX-7.

A reply would be greatly appreciated.

Thanks :-)

I'm interested in this comparison as well. Michael has partly implemented it already in respect of the noise comparison, which is a very important part of the overall story. But as Eric Chan correctly mentioned further above, the challenges must be greater putting 24 rather than 16 MP on the same size sensor. It is really of academic interest to me, because I already bought a 5N, I really like it, and it's highly unlikely I would upgrade to a 7 when it becomes available (see below). I think further comparison of other aspects of image quality apart from noise would be interesting, simply because it would provide further insight into how the technological frontier is being pushed in respect of outcomes from ever smaller pixel pitches.

The night before last was the first time I actually got down to making 13*19 inch prints with raw files from my 5N, shot with the 18~55 kit lens. I bought this camera because I've been doing a heap of really long-distance air travel lately and weight and portability became a consideration. So I was prepared for some trade-off between IQ and convenience, especially considering the price difference between a 5N and a Canon 1Ds3. But I have to say, it's quite amazingly good. Smoothness of tonal gradation is excellent, rendition of shadow detail at lower ISOs is excellent - in fact overall the DR seems great, sharpness of the kit lens over most of the image area is on the whole remarkably good (especially after mild capture and output sharpening), and I'm finding CA is less prevalent with this kit lens than I've observed with my Canon 24~105 L. There is some barrel distortion at wider angle settings, easily corrected in LR or ACR or PS. Using an Epson 4900 and Ilford GFS, I'm able to print at non-resampled resolutions in the range of 220~300 to a high standard of print quality. I think the print size I'm working with is at about the limit for this camera without resampling. If I wanted to make yet larger prints or have the luxury of more cropping headroom with an equivalent print size/resolution and no resampling, in that set of conditions a NEX 7 may be indicated. But personally I'm not feeling that need.

The two operational downsides I'd note with the 5n, although I think it is a truly well-designed and well-constructed camera system, is that the flash and EVF are a bit fiddly to screw-down correctly (probably just takes more getting accustomed to) and I find the movie button something I need to pay attention to avoid, but then again, there ain't much room on that little beast for putting stuff!.

[Isaac]

Douglas, could you please point me to how one does this. I have the NEX 5n and the EVF. The only instruction I could find is on page 143 of the 5n manual where they say the brightness of the finder can be varied by +/- 1, whatever that means, but nothing about contrast or the impact on the histogram.

I believe Douglas means lower the contrast setting for the current "Creative Style" - find "Creative Style" in the manual index and there should be further information there.

[EricV]

The methodology of comparing a cropped image from one camera to a full frame image from another camera almost certainly has a much larger effect on resolution than any difference in actual sensor quality.  What Michael has effectively shown is that a stitched image from the NEX-7, with enough stitching to enlarge the field of view to match the Leica, gives superior resolution.

[Mark D Segal]

I'm frankly having trouble seeing what's wrong with Michael's methodology. He took great pains with the focusing in terms of camera set-up and focusing technique. He used the same lens for both cameras, and one must assume that the Leica lens is at least perfectly (as possible) adpated to the Leica camera. Reducing the field of view of the Leica to what one sees from the NEX is necessary and doesn't change resolution because PPI doesn't get affected purely from cropping. There's just less image, the stuff that can't be compared being deleted. Downsampling the NEX resolution is necessary to standardize for that variable, so that's one difference of treatment that CAN affect relative image quality insofar as it introduces the question of what the resampling algorithm does to the quality of apparent image detail. We've been brought up to believe that all resampling comes at some sacrifice of IQ, depending on the quality of the algorithm, but downsampling generally less than upsampling, so if anything this could be expected to work against the NEX - but here too, Michael addresses the possible impact of that issue with the quote from Bob Newman - which if I've understood it in context means to say that if you start from a higher native resolution and downsize to the comparator lower resolution using a very good resampling algorithm, you'll still end-up with higher apparent resolution compared with that of the lower resolution camera - the point being that the resampling step doesn't vitiate the effort to compare apparent resolution of image detail.

[EricV]

I'm frankly having trouble seeing what's wrong with Michael's methodology ... Reducing the field of view of the Leica to what one sees from the NEX is necessary and doesn't change resolution ... There's just less image, the stuff that can't be compared being deleted. 

  Cropping is the problem.  More than half of the Leica pixels were thrown out at the start, given no opportunity to contribute to the camera comparison.

[bjanes]

  Cropping is the problem.  More than half of the Leica pixels were thrown out at the start, given no opportunity to contribute to the camera comparison.

+1.

[ErikKaffehr]

Hi!

The problem is that it is hard to say what question the comparison answers. The cameras in the setup have different field of view. A Sony NEX7 would not be a good replacement for a a Leica M9 with the same 50 mm lens. If we want to compare the two as picture making machines we would need to use lenses giving the same field of view on both cameras, with 35 mm lens being a good candidate on the NEX 7.

For the chosen crop the Leica would have 7.7 MP. In this comparison the Sony image is scaled to same size, so 68% of the resolution is effectively discarded. The image is downsized with bicubic sharper, which increases edge contrast and will also induce some aliasing.

What Michael's test shows is that the NEX 7 works well with a 50 mm lens and that the OLP filtering and the micro lenses don't lead to an unacceptable degradation of image quality.

Best regards
Erik

I'm frankly having trouble seeing what's wrong with Michael's methodology. He took great pains with the focusing in terms of camera set-up and focusing technique. He used the same lens for both cameras, and one must assume that the Leica lens is at least perfectly (as possible) adpated to the Leica camera. Reducing the field of view of the Leica to what one sees from the NEX is necessary and doesn't change resolution because PPI doesn't get affected purely from cropping. There's just less image, the stuff that can't be compared being deleted. Downsampling the NEX resolution is necessary to standardize for that variable, so that's one difference of treatment that CAN affect relative image quality insofar as it introduces the question of what the resampling algorithm does to the quality of apparent image detail. We've been brought up to believe that all resampling comes at some sacrifice of IQ, depending on the quality of the algorithm, but downsampling generally less than upsampling, so if anything this could be expected to work against the NEX - but here too, Michael addresses the possible impact of that issue with the quote from Bob Newman - which if I've understood it in context means to say that if you start from a higher native resolution and downsize to the comparator lower resolution using a very good resampling algorithm, you'll still end-up with higher apparent resolution compared with that of the lower resolution camera - the point being that the resampling step doesn't vitiate the effort to compare apparent resolution of image detail.

[bobtowery]

Michael's NEX-7 vs M9 comparison is badly screwed up, invalid, and thus, entirely pointless. [ET AL]

I find this reply completely over the top, unprofessional and unwarranted. At the beginning of the piece, Michael states that he loathes even posting these comparisons because people like you attack them.

It's fine for you to render your opinion on another way the test could be conducted.  There is no need to go nuclear, going on and on.  Replies like yours insure the rest of us won't get future information we can use from Michael. Does he need to put up with this crap? I doubt it.

Furthermore, his methodology is just one way, the way that worked for him. He clearly states there are many ways. No one is paying him to do this work, so I believe it is valid that he does it the way that suits him. Again, it's fine for the rest of us to suggest numbers of other methods, but really, what is in it for him?

This particular test methodology might not be the way many people would do it. Nevertheless, you can easily see there are many hours of effort, and if you read between the lines he is expending even more time to make sure there's no muff like his quick first attempt in San Diego. Even if you would do the test differently, there is good information here regarding the camera systems and the lenses. Free of charge, my friend.

Photography and its gear are subjective. Show me all the mtf charts you want: can you prove that a manufacturer's f/1.2 is worth 10x their equivalent f/1.8?

There was a thread recently about what websites compete with LuLa. Some other worthy sites were mentioned, but there is simply no other site that has this particular mix of content and expertise.  So I for one will thank you for not pooping in our punchbowl.

And lastly:

The days of the traditional rangefinder are over.

I wonder how long people have been saying this?

[Anadrol]

I think that there is no perfect methodology as simply sensor sizes are different.

Michael's test is interesting,
but it should also be done with a 35mm Leica on the NEX and a 50mm on the M9.
With if possible both lenses having about the same resolution.

Also, testing the Zeiss 24mm on the NEX VS a 35mm on the M9 would be interesting.
After all this would be comparing the best of what the NEX system has to offer VS the best of Leica.

I'm also curious about bokeh differences.

[ErikKaffehr]

Hi,

I agree entirely! Fortunately I'd think Michael has a thick skin.

Best regards
Erik

I find this reply completely over the top, unprofessional and unwarranted. At the beginning of the piece, Michael states that he loathes even posting these comparisons because people like you attack them.

It's fine for you to render your opinion on another way the test could be conducted.  There is no need to go nuclear, going on and on.  Replies like yours insure the rest of us won't get future information we can use from Michael. Does he need to put up with this crap? I doubt it.

Furthermore, his methodology is just one way, the way that worked for him. He clearly states there are many ways. No one is paying him to do this work, so I believe it is valid that he does it the way that suits him. Again, it's fine for the rest of us to suggest numbers of other methods, but really, what is in it for him?

This particular test methodology might not be the way many people would do it. Nevertheless, you can easily see there are many hours of effort, and if you read between the lines he is expending even more time to make sure there's no muff like his quick first attempt in San Diego. Even if you would do the test differently, there is good information here regarding the camera systems and the lenses. Free of charge, my friend.

Photography and its gear are subjective. Show me all the mtf charts you want: can you prove that a manufacturer's f/1.2 is worth 10x their equivalent f/1.8?

There was a thread recently about what websites compete with LuLa. Some other worthy sites were mentioned, but there is simply no other site that has this particular mix of content and expertise.  So I for one will thank you for not pooping in our punchbowl.

And lastly:
I wonder how long people have been saying this?

[Wayne Fox]

Since the images are being observed at 100% on screen, in effect pixel peeping is being done. No prints were made. In the real world, one would not crop the M8 18MP image to 7.7MP.

Regards,

Bill

Seems to me the M9 images in the article are at max resolution, because although he "cropped" them, he didn't resample them so the images are still at 100% as captured by the camera - they are as good as they can be.  The fact that he down sampled the NEX7 files to match the cropping seems quite similar to what would happen when printing.

[EricV]

The problem is that it is hard to say what question the comparison answers...

The test as conducted answers the question: does the full NEX-7 sensor (24MP) provide superior resolution to the center section (8MP) of the Leica M9 sensor?  The answer is obvious from a comparison of the pixel sizes -- the NEX-7 with 4um pixels has superior resolution to the Leica M9 with 6.8um pixels. 

Sensor size was deliberately excluded from the comparison.  Using the same methodology, a small sensor camera with 2um pixels would blow away either of these cameras, and a medium format sensor with 8um pixels would score worse than either of these cameras.

[michael]

Eric,

Yes, this is true in theory based on pixel pitch alone, but wouldn't be in practice, because the IQ would be crap. The whole point of this was that DxO has already passed their judgement on the relative merits of the sensors – excluding resolution. That's why the DxO chart was at the top of the segment.

I was then trying to level the playing field on the resolution topic. Guess some people just don't get it.

Michael

[EricV]

Seems to me the M9 images in the article are ... as good as they can be. 

They would be better if a longer focal length lens was used, so that all of the pixels on the sensor contributed to the image. 

[01af]

I'm frankly having trouble seeing what's wrong with Michael's methodology.

Then why don't you simply open your eyes and take a look? That's all it takes ...

He took great pains with the focusing in terms of camera set-up and focusing technique.

That's the primary point. He might have taken great pains in achieving accrate focus but as a matter of fact he ended up with inaccurate focus. Worse yet—and that's the reason why I am so furious about it—he didn't even notice when processing the files and carried on publishing the distorted result regardless.

Reducing the field of view of the Leica to what one sees from the NEX is necessary ...

It's necessary because he started with using the same lens on two different-format cameras. This method is fine when you're interested in comparing the same portion of the sensor—like, say, one square millimeter. But then why all this 'subjective analysis of real-world images yada yada' talk in the first paragraph? If he's interested in a result that correlates to real-world usage of these cameras then it would make sense to apply a methodology that will yield this kind of results. Shooting with the same lens does give valid results (provided focusing was accurate)—however not the kind of results he pretends to be interested in but a different kind. Both kinds are equally valid—but serve different intents.

But then—I keep repeating myself—the essential problem with this comparison is the inaccurate focus of the M9 shots. In relation to this, the methodology issue is secondary.

If we want to compare the two as picture making machines we would need to use lenses giving the same field of view on both cameras, with 35 mm lens being a good candidate on the NEX-7.

Exactly. When using the same lens and cropping out the same subject detail (as Michael did) then the pixel count for that detail with the NEX-7 will be almost 3.1× greater than with the M9. When using equivalent lenses (as a real-world photographer would do) then the NEX-7's pixel count for a given subject detail will be only 1.33× greater. So in Michael's test the NEX-7's advantage over the M9 appears 2.3× as big as it actually is. Or rather, it would appear 2.3× as big when focus would have been accurate. The M9's poor focus inflates the apparent difference even more.

What Michael's test shows is that the NEX-7 works well with a 50 mm lens and that the OLP filtering and the micro lenses don't lead to an unacceptable degradation of image quality.

Agreed.

[ErikKaffehr]

Hi,

Well, a lot of things happen when printing. Sharpening for output, dithering...

Anyhow, I don´t have an issue with pixel peeping. What's on the screen is what gets on the print. So if something is visible on screen it will also be visible in the print. But the differences are much larger on screen at actual pixels than on moderate size prints.

We had a recent discussion on diffraction and I was comparing a P45 image shot in the great MFDB shootout of 2006 at f/8 and at f/22, the difference caused by diffraction was as visible on the prints as on the screen. Printing in smaller size won't show the same difference. Sharpening matters a lot.

The enclosed image compares studio shots from "The Imaging Resource" for Nikon D3X and Pentax 645D, these center crops are flatbed scans from A2-size prints. The left and right image are sharpened using "Landscape" setting in Lightroom while the center one from the Nikon D3X is using deconvolution.

Best regards
Erik

Seems to me the M9 images in the article are at max resolution, because although he "cropped" them, he didn't resample them so the images are still at 100% as captured by the camera - they are as good as they can be.  The fact that he down sampled the NEX7 files to match the cropping seems quite similar to what would happen when printing.

[ErikKaffehr]

Hi,

You may be right on the focusing issue. On the other hand the Leica is not easy to focus exactly, at least according to Leica expert Erwin Puts and also Lloyd Chambers. There is also some probability that Michaels Leica is not aligned within 10 microns. Ten microns defocus would result in a circle of confusion of 7 microns a f/1.4.

Many testers assert correct focus by shooting focus brackets. But in real life you would probably not bracket for focus.

Another factor is that the Summilux with all probability has focus shift, the Summicron f/2.5 has a lot. That is actually a problem for both NEX and Leica. The NEX can be focused stopped down, but how do you adjust a coupled rangefinder? For correct focus at full aperture or for the lens stopped down to medium aperture?!

Best regards
Erik

But then—I keep repeating myself—the essential problem with this comparison is the inaccurate focus of the M9 shots. In relation to this, the methodology issue is secondary.