NEX-7 vs. M9 – Part Deux

[douglasf13]

...If you can live with the crop factor, the NEX-7 looks like a good alternative to the M9, using Leica lenses.

Best regards
Erik

Maybe in the center, but I was unimpressed with the NEX-7's corner performance compared to the M9 in Michael's first test, and we haven't seen corner examples from his second test.

[AlanG]

Nothing personal, but I think everyone else is "over thinking" it.

I did a screen grab of the full image from the Leica, and uprezzed that to the size of a Leica file ... @5200 pixels wide.  While very blurry, you can see that the size of the relevant information exactly matches the detail shots offered in the aricle (the second image is a screen grab of one of those).  The detail shots are at 100%, so the fact that he "cropped" the image doesn't seem to matter, because what he offers as detail shots are not resampled.

Sorry, but I can't follow what you are getting at or what images you are doing this to. I'm not sure if we are talking about the same thing.

From the math of the sensor size and pixel density it is very simple to see:

An APS-c sensor is 23.5 mm wide and on the Nex 7 this contains 6,000 pixels per line.
On an M9, a 23.5mm wide crop will only contain 3402 pixels per line.  3402=(23.5mm*5212 pixels)/36mm or (23.5/36)*5212

Assuming the same lens and subject distance was used on both cameras:
If you make a 100% crop from each (one image pixel per display pixel,) the Nex image will be 1.76 times larger in linear dimensions. This makes sense when you consider that the M9 only uses less than 8MP on the same area that the Nex 7 has 24MP.  In order to make them match in linear dimensions you'd either have to scale down the Nex image or scale up the Leica image.

In any case Part Deux answers the question by showing you need to shoot with a longer lens on the M9 to get the same number of pixels on a given area of the subject.

[Wayne Fox]

Sorry, but I can't follow what you are getting at or what images you are doing this to. I'm not sure if we are talking about the same thing.

I guess what I'm trying to say (poorly it seems) is that the math doesn't seem to jibe with what I'm seeing visually in the article.

So I tried to duplicate the original "files"  by taking screen grabs of the full image in the article from the m9 and from the area of that image that is from the NEX 7.  I uprezzed those to create files  to approximate dimensionally what came out of the camera (18mp for M9 and 24mp for NEX 7).  When I compare these to the side by side comparison shots in the article, the m9 detail of the m9 comparisons match up dimensionally with this "created" original I made. (the two images in my previous post).  Thus my believe that the m9 sections displayed in the article are at 100%, there was no resampling.  I also think any resampling of the M9 files would have been mentioned.  So the entire idea of cropping those files is irrelevant, because they just got cropped even further to show the small sections, but no resampling occurred.

The confusing part is with the NEX 7.  When I downrez my theoretical 24mp "original" file to 18mp, the resulting file detail does not match dimensionally at all the detail sections in the article... not even close.  To get a file size which matches the dimensionally the detail sections from the NEX 7, making up for both the higher rez sensor and crop factor, I have to down rez my 24mp "original" file to the equivalent of an 8mp camera, not 18mp.  (My rudimentary math skills also end up with this same number)  Maybe I'm missing something in the process as explained in the article ...

But agreed, the second test seems to address things quite well, so not much sense in all this.

[Johnny_Johnson]

Hi Alan,

We can only speculate since Michael hasn't addressed the question directly. However, in the review he states that he downsized the Nex image to 18 MP so that the images matched in both field of view and resolution. I think that the simplest and most likely answer is that he misspoke regarding the 18 MP and that he actually did nothing more than resample the Nex image down to match the cropped, 8 MP M9 image.

Cya,
Johnny

[AlanG]

Hi Alan,

We can only speculate since Michael hasn't addressed the question directly. However, in the review he states that he downsized the Nex image to 18 MP so that the images matched in both field of view and resolution. I think that the simplest and most likely answer is that he misspoke regarding the 18 MP and that he actually did nothing more than resample the Nex image down to match the cropped, 8 MP M9 image.

Cya,
Johnny

That makes complete sense but then it "probably" would not have shown significantly more detail than the Leica M9 shot. (That's another test in itself and not what he said he was looking for.)   Either one was scaled up, one was scaled down, or both were scaled. (So there are 3 possibilities of what happened.)

In any case, we don't really know what he did in that test but we do know what the cameras can do.  And I'm ordering a Nex 7 despite not really needing one. I tried it out at the Photo Plus Expo and was very impressed by the handling. Now, after seeing the test results here and elsewhere, I am really looking forward to use it for something...?  I'm grateful for his thorough review and tests despite my questioning this one aspect.

[ErikKaffehr]

Hi,

Regarding the first test, the corners are pretty clean on both images at f/5.6, so it looks to me that the softness we see at f/1.4 and f/2.8 is caused by the lens.

The Leica image has a somewhat cooler rendition and possibly higher contrast, which can affect our impression of sharpness. This is one of the perils of testing, there are many. Including a Color Checker in the image and adjusting color balance and contrast in rendition can make the visual comparison more exact.

A small observation, with fast lenses we often have a wavy field curvature. The image plane of the lens is not flat but shifts a little back and forth. This can be seen in the MTF chart for the Summilux 50/1.4 at full aperture.



Stopped down to f/5.6 the field is flat.



The images are measured MTF from this article by Erwin Puts: http://www.imx.nl/photo/technique/technique/hslenses.html

This is a case where the APS-C sensor actually doesn't utilize the "sweet spot" of the lens but actually uses a part of the "weak spot".

Regarding the MTF curves, keep in mind that they go only to 40 lp/mm while the sensor pitch on the NEX7 corresponds to 119 lp/mm. MTF-plots for the new Schneider and Rodenstock digital lenses take this into account and show MTF for 80 lp/mm if I recall correctly.

Check this short discussion by our ambitious friend "bjanes" http://www.luminous-landscape.com/forum/index.php?topic=60353.msg486958#msg486958 .


Best regards
Erik

Maybe in the center, but I was unimpressed with the NEX-7's corner performance compared to the M9 in Michael's first test, and we haven't seen corner examples from his second test.

[AlanG]

Hi,

Regarding the first test, the corners are pretty clean on both images at f/5.6, so it looks to me that the softness we see at f/1.4 and f/2.8 is caused by the lens.


Best regards
Erik

It could be the lens, but also consider there is a lens adapter in use too. We really don't know if the lens was perfectly square to the sensor.

[Ray]

Corner performance of lenses designed for the cropped format can be a very noticeably problematic for those of us who have got used to the enhanced corner performance we get when using full-frame lenses on a cropped-format camera, as I have.

Out of curiosity, I checked out the Imatest results at Photozone for the Sony E-mount 18-200mm, which looks like a very useful lens for the NEX-7.

Photozone used a 16mp NEX-5 for the test, so maximum resolution in the centre should be greater with the NEX-7. Unfortunately, resolution in the corners of the frame is abysmal for this lens. A higher resolution sensor will not change this fact.

Of course, corner resolution is much improved at F8 and F11, as is always the case. My point is, it seems a shame to buy into a new system which requires new lenses that are not on a par with what one already uses.

The 18-55/F3.5-5.6 kit lens for the NEX cameras seems to have better edge and corner performance than the 18-200mm, but still not as good as the Canon equivalent 17-55/F2.8, as tested on the 15mp Canon 50D.

Not only is this Canon lens also designed for the cropped format, and therefore relatively light, it has a constant F2.8 maximum aperture across the whole FL range, in addition to significantly better edge and corner resolution.

Even at F2.8, corner performance is better with the Canon lens than at any apertures with the Sony lens, except at F8. The Sony lens has very marginally better corner resolution at F8 than the Canon lens at F2.8, on average. But only on average. The best corner resolution for the Canon 17-55 at F2.8 is equal to the best corner resolution for the Sony at F8. That is, 2135 'Line Widths per Picture Height' for the Sony at 55mm and F8, as opposed to 2138 LW/PH for the Canon at 24mm and F2.8.

Perhaps it needs to be mentioned again that the NEX-5, in addition to being a more modern camera than the Canon 50D, has a slightly larger sensor (1.5x crop factor as opposed to 1.6x for the Canon), and slightly more pixels. The NEX-5 may also have a slightly weaker AA filter which would explain the slightly higher resolution in the centre of the Sony lens.

Bottom line, I would feel very foolish if I were to order an NEX-7 with 18-55mm kit lens if I later discover that Canon announces  a similar design of camera next year, also with 24mp, which I could use with my existing 17-55/F2.8, and which although a slightly heavier lens than the Sony 18-55, is better and faster.

There's a lot to be said for choosing your lenses first, then finding a body to suit.

[douglasf13]

Another wrinkle in the whole equation, Ray, is that the NEX-C3 and NEX-5N perform better in the corners with rangefinder wides than the older NEX cameras and the new NEX-7.  It is certainly the case with corner color shift, and the evidence is starting to show it to be the case with resolution as well.  It seems Sony did something a little different with the sensor filter or micro lenses on the 5N and C3's sensor.

[ErikKaffehr]

Hi,

Good point. On the other hand sharpness on both Leica and Sony is pretty bad. Also there is a lot of lateral chromatic aberration on both. The "corner" image lays pretty much in the weak spot of the MTF-chart. These other factors point at lens being the issue. At f/5.6 the NEX-7 is possibly little bit better, but at that aperture the MTF diagram is also very good.

The NEX-7 could make use of MTF up to 110 lp/mm, but no MTF chart I have seen goes that far.

Best regards
Erik

It could be the lens, but also consider there is a lens adapter in use too. We really don't know if the lens was perfectly square to the sensor.

[Maurício Costa]

Well, thank you Michael and sorry for starting the mess...

Anyway the methodology now is perfect (though there is no way to perfectly compare the two cameras) - it couldn't be better.

It showed clearly now that there is not much of a difference (I saw more difference in the first test), though the NEX gets an advantage.

Interesting the moiré comment....

Another thing that matters me today when evaluating a camera is the esthetics of the noise. For large printing, the noise patterns are for me as important as the grain quality was at the film age... it would be nice to see that evaluated by Michael in further tests.

[ErikKaffehr]

Hi,

Michael did make a noise test, and the samples at DPReviews demonstrate noise.

Best regards
Erik

Well, thank you Michael and sorry for starting the mess...

Anyway the methodology now is perfect (though there is no way to perfectly compare the two cameras) - it couldn't be better.

It showed clearly now that there is not much of a difference (I saw more difference in the first test), though the NEX gets an advantage.

Interesting the moiré comment....

Another thing that matters me today when evaluating a camera is the esthetics of the noise. For large printing, the noise patterns are for me as important as the grain quality was at the film age... it would be nice to see that evaluated by Michael in further tests.

[douglasf13]

Erik, did you happen to read the Zeiss wide angle paper that came out a few weeks ago?  With more symmetrical lens designs, it's pretty incredible how much the sensor filters affect mtf:

[Maurício Costa]

Yes, the noise test is there, but I would like to hear more from him of his subjective impressions regarding different noise patterns the cameras produce.

[ErikKaffehr]

Hi!

I was aware of that effect but I missed it in Zeiss publication. Leica M5 had a spot meter cell swinging out of the optical path before exposure, so I think most newer Leica lenses has more clearance behind the lens to make room for that sensor. That would suggest  that Leica lenses would work better than traditional Biogons.

It is my understanding that modern lenses take assembly in front of the sensor into account. When IR filter is removed it is replaced with optical glass of the same thickness.

Let's not forget that very good retrofocus lenses can be built. The Zeiss 21/2.8 is an awesome lens and so is the Nikon 14-24/2.8. But both lenses are big!

Best regards
Erik

Erik, did you happen to read the Zeiss wide angle paper that came out a few weeks ago?  With more symmetrical lens designs, it's pretty incredible how much the sensor filters affect mtf: