In a comment to the previous post, a reader challenged my assertion that the NEX-7 is a terrible lowlight camera because of high noise level above ISO 400. I’ve conducted some experiments to demonstrate what I’m talking about.
Because noise needs to be evaluated at pixel for pixel or higher resolutions, I thought it would be best to compare the NEX-7 to another 24 megapixel camera. Fortunately, I had one available: the Nikon D3x. The D3x’s low light problems are well-known. Even Nikon acknowledges them tacitly by using identifiers such as “H 1.3” for settings above ISO 1600.
As you will see, the NEX-7 has low light performance that is inferior to the D3x. At one level, this is to be expected, since the D3x is a full frame camera, and therefore has sensor sites with approximately 1.5 [correction: 2.25] times the area of the sensor sites on the NEX-7. On another level, it is a surprise, since it’s been almost 3 years since the D3x came out, and technology marches on.
On to the test. I put a Leica 24 mm f/3.8 Elmar M ASPH on the NEX-7. Multiplying 24 mm times 1 1/2 gives 36 mm for the focal length with the equivalent field of view on a full frame camera. I didn’t have a 36 mm lens, but I did have an AF-S Nikkor 35mm f/1.4 G. I put it on the D3x. Serially, I aimed the cameras at a dimly lit bookcase, set the ISO to 800 (on the NEX-7, 800 is the next ISO above 400), focused with live view, set the aperture to f/8, and the shutter speed to an eighth of a second.
I converted the images from raw format in Lightroom 3.6, turning off all noise reduction, and leaving sharpening at the default.
Here’s the picture with the NEX-7:
Here’s the one with the D3x:
In order to look closely at the noise, I expanded the images to a magnification of three, using nearest neighbor, which will give a square array of nine identical pixels for each pixel in the original images.
Here’s the blowup of the NEX-7 image:
Here’s the expansion of the D3x image:
You can see that the NEX-7 has more noise in general, that the noise patterns appear to be coarser, and that there’s more color noise. These differences are more apparent before the JPEG compression that I used to post the images on the blog. I always learn something from reader’s criticisms, and I retract the word “terrible” in the previous post. Let’s say that, with cameras like the D3s available, the NEX-7 is far from the best choice for a low light camera.
The NEX-7 image appears to be the same or maybe slightly sharper than the D3x photograph, even though the sensor is smaller. It’s very close, though. It looks like the Leica glass is able to overcome the Nikon’s sensor size advantage. Also note the greater color fringing in the Nikon image. Sure makes the NEX-7 seem like a bargain, even if the Leica lenses aren’t.
Alain says
What an insteresting comparison between a FF camera and an APSC …. ???
So with your pixel peeping review we have to admit that DXO mark and Dpreview are making a huge mistake saying that the Nex 7 controlled high iso very well up to 6400 ISO !!Dpreview :
Image Quality
“The NEX-7’s image quality is difficult to fault. The 24MP sensor is capable of recording huge amounts of detail (just as long as your lens can deliver it), while also offering excellent high ISO performance for low light work, with quite useable results up to ISO 6400”
“Compared to the Micro Four Thirds flagships – the Panasonic Lumix DMC-GH2 and the Olympus PEN E-PL3 – the NEX-7 performs visibly better at the very highest ISOs, showing the advantage of its larger sensor. This becomes very apparent at ISO 3200 and higher, at which point the E-P3 and GH2 are both struggling to suppress noise while retaining colour saturation and low-contrast detail”
Jim says
I believe the difference between the reviews you cited and this blog are due to using different standards of comparison. The reviews you mentioned compared the NEX-7 to cameras with smaller sensors. I’m comparing it to other cameras you could use for low light situations, regardless of sensor size.
Because of its poor low-light performance compared to the D3 and D3s (and the slower speed and the fewer images that it can buffer), I consider the D3x a special-purpose camera, for use only when I need the resolution. I looked at the exposure counts on my two D3s’s, and they total over a hundred thousand, which is twenty times the exposure count on my D3x, even though that camera is older.
The NEX-7 has even poorer low-light performance than the D3x. That means that I’m certainly not going to pick it up when I head into a low-light situation, unless portability or its unique live focusing are dominant criteria for the project.
None of this is to say that the NEX-7 is a bad camera, or even that its low light performance is poor given its sensor size and pixel count; given those constraints, it seems to do a good job with noise.
I am working on a post about dealing with noise in hyper-resolution cameras.
Jim
Peter says
I have to agree with Alain.
Dpreview made comparison, which has more sence than your point of view.
I’d say, that NEX should be compared with m4/3 camareas and not with professional camara D3s (which cost 5X more than NEX 7). We can compare it (similar size of chip), but I would never say that NEX7 is bad camera for low light.
I would say that old professional camera with 4Mpx will beat “all” compact cameras (~>12Mpx) 2011…
I can agree that NEX 7 could have beter performance, if SONY would use 16Mpx chip (NEX5). The NEX7 is still quite very good in low light when you compare with the same class of cameras.
I’m just trying to decide, which camera will be best for me. I’m thinking about NEX7, GX1 and PL3 for now. I like m4/3 for lenses and faster AF, NEX7 has internal EVF and is more complex from my point of view. Maybe there will be released a better camera from SAMSUNG (better than NX200) or Fuji during this year. We will see but I have to buy new camera in 11 month…
BR Petr
P.S. Sorry for my English
Jim says
Peter,
First off, you should consider the context of my original remark. We have a new kind of camera in the NEX-7: small size, big sensor for the size, eye level electronic viewfinder. What kind of photography is that camera good for? I was not trying to compare the NEX-7 to other, in-some-sense-similar, cameras. I was, and am, more interested in the properties of the class of cameras of which the NEX-7 is the harbinger.
If you’ve already decided that you’re going to buy a camera kind of like the NEX-7, then my way of looking at what the camera is good for is pretty useless (you’ve already made up your mind), and the dpreview comparison is much more appropriate.
You brought up the issue of money. I was ignoring the cost of the camera bodies. I think that mindset comes with hanging Leica lenses on the NEX-5 or 7.
Jim
PS. Your English is a whole lot better than my Czech.
Simon says
I don’t have a test to prove anything (though I dont think much is proven by comparing sensors of different sizes) but I do think a sweeping assertion that that the NEX 7 is terrible in low light is more than a little harsh.
Anecdotally, I can say I haven’t noticed a difference from my NEX 5N (and please don’t try and say the NEX 5n has poor low light performance) on my 27″ iMac screen until I go beyond ISO 1600. Even then I get usable results for prints or on the web up to Iso 6400.
If your version is so poor I would suggest it is defective.
Jim says
The word “terrible” was retracted in the this post. I left it in the previous post because I don’t like to change posts arbitrarily; I think it’s confusing. That said, I will make the correction there, since not everyone will read both posts.
What noise is acceptable depends strongly on what your end use of the image is. For full-res images printed at 360 ppi (17 x 11 in) on my Epson 4900, I find the NEX-7 noise at ISO 800 to be obtrusive in low-spatial-frequency areas of the image, and I need to take post-processing steps to deal with it. More on these steps in a future post.
I doubt if my NEX-7 is defective in this regard, since the noise observed bears the relationship to the D3x noise that one would expect with sensors of similar technology, although I had hoped that three years had improved things.
Jim
John says
Interesting that you are comparing the NEX-7 with a full frame sensor camera. There is nothing wrong with this as you openly declare this and that the D3 pixels are 1.5 times bigger and hence the improved noise performance.
Why I say it is interesting is that you should get a similar improved noise performance in the NEX-6 that also has pixels 1.5 times the size. Of course this is because it has 16.1 megapixels rather than 24 megapixels. However, if the photography you do is suitable (for myself and many others), than maybe the NEX-6 is a highly suitable alternative.
What is the opinion of everyone else on this ?
John says
I do beg your pardon, the comparison is not quite the same; the NEX-6 is 1.5 times the area per pixel whereas the D3 pixel is 1.5 times the size in each linear dimension. Hence the D3 pixel is actually 2.25 times the size in area (assuming similar quality of sensor design – and it is Sony that supplies most of the sensors 🙂 ).
It is the area that actually matters as this sets the number of photons captured in the pixel (all other things being equal). However, the argument still remains that the NEX-6 should be a significantly better noise performer than the NEX-7.
The NEX-7 has good noise reduction I believe. However, as I have learned from writing image processing algorithms, you pay the price as information will be lost in the process and something somewhere will be degraded. Moreover, there is no reason why noise reduction cannot be applied in external post processing if wanted.
Jim says
John, I don’t disagree with any of that. I was using the D3s and D3x as examples of sensors of identical size, similar technology and different resolution to draw a conclusion about the NEX-6 and NEX-7.
I’m sure there are theoretical losses in downresing. I’m sure that there are practical ones in some cases. That said, I haven’t seen major problems in downresing D3x or NEX-7 images.
Have a look at this: http://blog.kasson.com/?p=1362 and http://blog.kasson.com/?p=1369 and http://blog.kasson.com/?p=1377 and http://blog.kasson.com/?p=1385 and especially http://blog.kasson.com/?p=1391 and http://blog.kasson.com/?p=1405
Jim
Jim says
John, thanks for the correction about the D3s vs NEX-7 capture area. I’ve fixed it.
Jim
Jim says
John,
SInce the publication of this post, I’ve determined that I can get a D3x to give similar noise performance to a D3s if I res the D3x image down to D3s resolution. Indeed, to a first order, I now believe that for sensors of similar technology, noise at a given resolution is a function only of sensor size. If that is the case, the NEX-7 is a more versatile camera than the NEX-6, in that it can deliver high res at moderate noise, or, res’ed down to 16 Mpixels, lower res at lower noise.
That’s my theory, anyway; I don’t have an NEX-6 to use to test that theory.
By the way, noise and exposure are related; see this post: http://blog.kasson.com/?page_id=2457
Jim
John says
Jim,
Are you suggesting reducing the resolution of the NEX-7 by some software technique within the camera ? If so, this certainly will not reduce the fundamental noise of the sensor. Signal to noise ratio is set by the number of photons received in each pixel; it is a quantum effect caused by uncertainty in the arrival time of photons. All you are doing by reducing the the effective number of pixels by a software technique is to average the noise locally.
This is why Sony produces sensors of lower resolution of the same frame format because they know this is the only way of getting better fundamental noise figures. One must pay ones money and take ones choice.
My feeling is that the best customer choice is to work out what resolution you really need, not what the camera manufacturers can offer. This should take into account, fundamental photographs, the increased demands of photo printing and the likely enlargements you want to do. I find that 12 megapixels is OK for most photographers – including myself. However, some photographers may want more. Once chosen, go for a frame format sensor that is large as your budget sensibly accommodates bearing in mind that you want lenses and other things as well. Of course, the mix is complicated by lens characteristics, functionality of the camera, the list goes on and on 🙂
I like your original post in that you have demonstrated well the trade off of pixel size and noise and it has raised interest in others. Interesting to talk to you.
John
Jim says
John,
I actually think the best way to go in the future is to raise the pixel density well beyond the resolving power of the lens, drop the AA filter, and res the image down to what meets the noise and resolution needs of the photographer. We’re not there yet, but another two doublings of the D800E reolution will put us there.
It’ll be interesting to see how it all turns out.
And, no to the in-camera processing. I think that this is all best done in post.
Jim