I noticed this before, but did not have the lenses you tested. now you published the results for the 85mm, which I do own. What is surprising me is the low center resolution for this lens (and for all your NEX-7 results). In practice I see a camera that has a higher resolution than the Sony a900.
I have made a simple comparison; a test chart (printed approximately A2, so not comparable with standard test chart), at such a distance that the test chart fills, horizontally, approximately 50% of the image.I changed the distance to make the subject scale the same. developed RAW's in LR4, with the same sharpening (amount 60, radius 0.5, detail 50 and masking 0). I had to change exposure a bit of the a900, since it gave, for the same f stop and shutter speed, a darker image.
717 is the NEX-7 and 2 is the a900.
It is clear that the NEX-7 has a light AA filter. there are a lot of color artifacts, including the colorization around black edges. and to me, the NEX-7 images seem to have a higher resolution.
Is it just that Imatest interpreted the results differently? Am I doing something wrong?
I know there are slightly more sharpening artifacts in the NEX-7 image, something I blame on the image being sharper to start with.
the comparison is show at 200%
Posts: 8,004
Threads: 1,851
Joined: Apr 2010
Reputation:
49
03-24-2012, 05:06 PM
(This post was last modified: 03-24-2012, 07:35 PM by mst.)
The results are not cross-system comparable.
We always said that and we will continue to do so.
The used RAW converters between the two systems are already different. Just to mention an obvious point here.
Technically I agree - the NEX 7 has a weaker AA filter.
However, if I applied more aggressive settings to the conversion profiles there would have been a danger to land beyond Nyquist at times and, frankly, I'm tired of that discussion.
I can Imagine you want to be careful with sharpening. However, now it seems as if the NEX-7 is the worst of the high resolution cameras. For the Nikon D3x the maximum is 4000, for the Canon 5DII the max is 3700, for Sony FF the max is 3600. And if I remember correctly, you have been commenting on the rather strong AA filter of the a900, compared to, for example, the 5DII .
You have been rather negative about the NEX-7, and I can see where you are coming from. But one of the aspects where it really shines, pure resolution, you have played down the performance to avoid certain a discussion that, in itself, has nothing to do with the camera but with the way it is tested.
And although it has been stated many times in the past that results are not cross platform comparable, I doubt many people looking at the results are aware of that. I would expect most people visiting your site will look at the test result and never read the forum.
Is it not be possible, that when you showed more comparable resolution figures, it would also become clear that one of the problems with such a camera is the enormous resolution potential, asking a lot of even the best lenses?
Posts: 8,004
Threads: 1,851
Joined: Apr 2010
Reputation:
49
03-25-2012, 08:13 AM
(This post was last modified: 03-25-2012, 08:20 AM by Klaus.)
The results are not cross system comparable.
If this simple sentence is not understood we can't help it anymore.
The NEX 7 has no AA filter on the horizontal axis. At the current settings the LW/PHs are at 3800 here - thus close to Nyquist. If I chose more aggressive setting this value would have been WAY beyond Nyquist. We can argue that Imatest can look a bit beyond Nyquist (although I don't really want to argue here anymore) but there are limits to this. Thus 3800 is very reasonable here.
The filtering on the vertical axis is stronger. On the average we are at 3400.
Canon and Nikon have a symmetrical AA filter so there's no real risk to end way beyond Nyquist here.
Posts: 621
Threads: 46
Joined: Apr 2010
Reputation:
0
[quote name='Klaus' timestamp='1332608815' post='17001']
...
However, if I applied more aggressive settings to the conversion profiles there would have been a danger to land beyond Nyquist at times and, frankly, I'm tired of that discussion.
[/quote]
Why not? We haven't had that for a long time. I really miss it <img src='http://forum.photozone.de/public/style_emoticons/<#EMO_DIR#>/tongue.gif' class='bbc_emoticon' alt='  ' />
You might remember the paper I found recently and shared here in the user forum: oatao.univ-toulouse.fr/303/1/estribeau_303.pdf If you look at figure 6 and the texts around it, it explains nicely the ideas behind the "slanted edge" method for measuring MTF. If you look at the figure it explains how you get a better sampling of the edge than the sensor pixel pitch would suggest. One is not violating Nyquist's theorem here at all - by combining the results from several lines, you use a higher sampling frequency than the pixel pitch, that is all. Anyway all Bayer sensors (try to) do that anyway. The pixel pitch of the green sites is by sqrt(2) larger than commonly quoted.
If one looks at oatao.univ-toulouse.fr/297/1/estribeau_297.pdf by the same authors, figure 16 shows examples of how strong a signal beyond Nyquist can be obtained with "slanted edge" methods.
Figure 16 of the second paper also shows, when using a proper calibrated slanted edge method, on can obtain results that are essentially compatible to traditional sine-pattern methods. For sure your results will not be, since the cameras you use as a sensor are not calibrated. You know and say that every time this topic come up.
All in all this is quite interesting and new to me. Once one understands it, it is just so obvious what is going on here <img src='http://forum.photozone.de/public/style_emoticons/<#EMO_DIR#>/smile.gif' class='bbc_emoticon' alt='  ' />
Have a good evening.
Joachim
enjoy
[quote name='Klaus' timestamp='1332663195' post='17023']
The results are not cross system comparable.
If this simple sentence is not understood we can't help it anymore.
The NEX 7 has no AA filter on the horizontal axis. At the current settings the LW/PHs are at 3800 here - thus close to Nyquist. If I chose more aggressive setting this value would have been WAY beyond Nyquist. We can argue that Imatest can look a bit beyond Nyquist (although I don't really want to argue here anymore) but there are limits to this. Thus 3800 is very reasonable here.
The filtering on the vertical axis is stronger. On the average we are at 3400.
Canon and Nikon have a symmetrical AA filter so there's no real risk to end way beyond Nyquist here.
[/quote]
With what lens did you reach that 3400 figure? I have not seen anything that goes beyond 3100 yet, which seems kind of ridiculous. Especially since I see clear aliasing with the CZ85mm. And although I get your reason for being careful, it still seems like you push the performance of all the lenses down for this.
Posts: 8,004
Threads: 1,851
Joined: Apr 2010
Reputation:
49
03-27-2012, 11:11 AM
(This post was last modified: 03-27-2012, 11:12 AM by Klaus.)
[quote name='waardij' timestamp='1332845182' post='17088']
With what lens did you reach that 3400 figure? I have not seen anything that goes beyond 3100 yet, which seems kind of ridiculous. Especially since I see clear aliasing with the CZ85mm. And although I get your reason for being careful, it still seems like you push the performance of all the lenses down for this.
[/quote]
E 16mm: 3281 LW/PH
E 30mm: 3392 LW/PH
A 16-50mm: 3399 LW/PH
If it helps - I was also surprised that the 50/1.8 and 85/1.4 didn't reach "the roof".
Ok, I missed that. Sorry. Let me ask one last question; did you ever try if the relation stays linear? Linear in the sense that when you change the sharpening, the relation between centre, edge and corner stay’s the same? If this is so, for me the conclusion is that the poor results on many e- lenses is indeed because 24Mp is asking a lot on aps-c, which it is.
For myself I have mostly looked at DSLR lenses and of those only really good ones combine with this camera do give magic.
Maybe also of interest; Lloyd Chambers, of Diglloyd.com, stated recently that he prefers the IQ of the NEX-7 over the first results he got out of the new Canon 5DIII (low iso). Confirming (for me) that the NEX-7, with the best glass, can be very good. He was not primarily talking resolution here.
Sorry if I go on a bit about this, it is just that it seems so strange that a camera that gives me such wonderful results is scoring so poorly in your test. Does not change anything to my results though.
Posts: 8,004
Threads: 1,851
Joined: Apr 2010
Reputation:
49
Yes, I did, of course, play around with different setups including one with a max. LW/PH at 3800.
As you surely know the sharpening effect is depending on the contrast of an "edge" on pixel level. Thus sharpening will always emphasize edges that are already sharp anyway. This is espeically "dangerous" regarding the lack of AA filtering on the horizontal axis. A sharpening of a "clean" edge on this axis results in a fairly extreme boost. Imatest does detect this and provides a figure defining the "over-/under-sharpening" of an edge. This figure is a key calibration point.
A higher sharpening (than the one currently used) does there always emphasize the center - less so the less-sharp corners. The "spread" between center and borders/corners would have been higher (LW/PH-wise).
However, we would have also used a different scale, of course. The "current" borders/corners are very slightly "better" (visually on the chart) compared to the version that I would have used with a more aggressive sharpening. However, the visual difference isn't substantial (and in fact it's less than I anticipated).
Tanks for that. Sort of expected you would have tried.
On sharpening; most sharpening works like that. Especially if masking is used, since this suppresses sharpening of low contrast edges. However, if deconvolution sharpening is used, and the point spread function is estimated correctly, the sharpening should do no more than compensate for the spatial low pass filtering that is caused by the lens itself, diffraction, the AA filter etc. MTF 50 is only one measure for the resolution of a lens. When the SNR of the file is good enough, a lower MTF, for example MTF 15, can also be very interesting in that contrast for detail with such a high spatial frequency can be restored. And as you will know better than I, the relation between MTF-50 and MTF-15 is not the same for each lens.
|
