raptor3x said:
jrista said:
The NX1 hit that noise threshold at a higher ISO than the 7D II.
That's not actually true, the reason the NX1 has a higher score is because the 7D2 hits the color sensitivity threshold at a lower ISO. The ISO score is composed of the minimum of some threshold value for dynamic range, SNR, and color sensitivity.
Right...and the 7D II loses color sensitivity because of it's higher color noise. It's also got Canon's characteristic heavy red cast (and I honestly don't know what causes that). It all boils down to noise, one way or the other. The NX1 has extremely low color noise, which is probably why it has such rich blacks.
Canon cameras also have the strong bias signal. I have actually learned through my astro processing that much of the larger scale noise and blotch frequencies that you see when you heavily stretch (shadow push) Canon data is actually from the bias signal. There are ways of dealing with that. You probably wouldn't really gain DR (not in the sense of gaining more data deeper in the signal to push), but you can improve Canon's signal characteristics by subtracting the bias signal away. That can remove a lot of ugly junk from deep shadows at low ISO, or at very high ISO (3200 and up). The potential consequence is that you might be throwing away some useful detail...kind of like Nikon's black point clipping.
raptor3x said:
jrista said:
That has always been the case with high ISO performance. Beyond somewhere between ISO 800 and 1600, cameras become physics limited. Read noise levels drop to fairly common minimums (somewhere between 1-3 electrons, which is basically meaningless under normal use situations), and SNR becomes a much more significant factor than dynamic range most of the time.
I should have clarified in the previous post, but the whole point I'm trying to make it that we don't really know how much of a difference the BSI sensor is actually making. I completely agree with you that for high ISO values the camera performance is going to be completely dominated by sensor area and quantum efficiency. The NX1 has a slightly larger sensor, so if BSI was as beneficial as you were suggesting then we would expect to see the NX1 beat the 7D2 in at least SNR by some significant margin. Instead what we see is that SNR ratio is actually higher in the 7D2 at high ISO (my guess is that this is due to a weaker CFA.) Comparing the NX1 to a D7200 or A6000, you'll see that the NX1 is still very slightly behind, or at least not ahead. I think BSI will become much more important as pixel densities continue to increase but I'm not sure how much improvement it will actually provide on a much less dense 42MP FF sensor, although I'd really love to be completely wrong on this one.
I understand. There
are different kinds of noise. The NX1 seems to have higher random (Gaussian) noise levels...however it has extremely low color noise levels, no banding to speak of. The higher random noise probably hurts it on the SNR front, but it's clean, pure random noise...that is EASY to take care of (and, therefor, easy to greatly improve the SNR of an NX1 RAW, possibly quite considerably, with careful NR techniques). The 7D II may have slightly lower random noise levels, but it's got the same old color noise issues that every Canon camera has. That is what hurts it's color accuracy. I suspect it is because of weaker CFA, but there could be other reasons (Canon's sensor tech is still quite archaic by today's standards). The bias signal has the same banding issues that every Canon camera has. That may not be properly taken into account by DXO's measurements (or anyone;s measurements for that matter)...but it absolutely affects aesthetic.
Just look at high ISO images from the NX1 and compare them to the 7D II. You should see what I am referring to regarding color neutrality and rich blacks. There IS noise there...but it's the overall aesthetic. I'd take the NX1 over the 7D II every time...because of the noise characteristics, as well as the high frame rate. The loss in bit depth at 15fps doesn't bother me...usually photon shot noise will diminish DR below 12 stops at high ISO anyway, so it's not an issue. And 12 stops is more than any Canon camera has ever achieved at native size as well.
raptor3x said:
jrista said:
There is also something about the NX1 color balance that I find very intriguing. Most cameras, including Nikon's and Sony's have a slight color cast, usually towards red, when you get to very high ISO settings or dig really, really deep into the signal. The NX1? It has the richest, deepest, and most neutral blacks I think I've ever seen. Noise levels are not as low as some cameras on the market...but the data looks so much better despite that. Plus, the noise characteristic is worlds better than the 7D II, which still suffers from Canon's bias signal patterns. It is very easy to reduce clean, random noise, which the NX1 has, and more difficult to clean up the blotchy, banded bias signal. One option with the 7D II is to create a master bias frame, and subtract it. That can improve things a fair bit...but, it may also clip some information if you are not careful about how you perform the subtraction.
Did you ever get the NX1? I remember the last time we talked you were planning to rent one but I didn't really pay attention after that. I only have the DPReview samples to go by but what I've seen is that the NX1 tends to have a much purer blue cast to the shadows while Canon and Sony sensors have much more of a reddish cast, so we may be seeing the same thing.
No, have not purchased one yet. I found the A6000, and that's changed my plans. I am actually waiting for the A6100 to hit before I decide between the three. Both brands deliver high frame rate and high IQ in compact, light weight, portable packages. I really LOVE the NX1 concept...but, there is lens compatibility to consider. The A6000 would be compatible with A7r II lenses...I have to consider that.
Regarding color cast...I have never seen any color cast in the NX1 at all. No blue, no red, no green. It's just dead neutral black. My screen is properly calibrated. I guess I could look on some other screens, but that neutrality in tone, the lack of color cast, is part of Samsung's technology. The ultra short distance between microlens and photodiode nearly eliminates color crosstalk, which is a big source of color noise. The only thing they could do better is actually use their ISOCELL technology, which would completely eliminate all color crosstalk entirely. I'm not sure that would improve things enough to matter, though...as the current NX1 sensor has the purest color I've ever seen (better even than Sony sensors, which still seem to have a faint red cast, or maybe a slight purple cast with Sony cameras.)
raptor3x said:
As for the blotchy behavior, that's definitely there but I've found it to be less of an issue with non-Adobe raw converters and not completely unique to Canon. My A7R exhibits similar behavior but the spatial frequencies of the blotches are a bit different.
I agree here. I actually try to use VNG demosaicing with Canon data. It seems to handle it MUCH better than AHD, which is used by Lightroom/ACR. I think it may actually be Adobe's actual implementation of AHD even. I was poking around with PixInsight recently...I found it actually does have the option of loading images with AHD, and when I did, while I found a little bit of banding in a couple places in some images, the results were FAR cleaner than Lightroom/ACR has ever produced.
That is rather sad. So many people use Adobe products, the fact that their implementation of AHD has such a detrimental impact on Canon data is a travesty. That said, read noise is read noise, and the blotch doesn't disappear. It takes on a better characteristic with VNG demosaicing, but the read noise levels are still high, and I've never seen any significant improvement in dynamic range...nothing that would ever allow me to stretch a Canon RAW as much as I can stretch Sony or Nikon RAW.
I have seen, on a couple occasions, some faint purple blotch with the A7r. However, I had to go beyond 6 stops into the bowels of the signal to actually encounter it. I've pulled data up as much as 8 stops, and that seems to be about the limit before you finally end up buried in Sony's read noise and/or bias signal. (I think the purple blotch is actually in the bias signal.) I don't need to stretch data that much in the vast majority of my work...the only time I ever actually NEED to do that is with astrophotography, but I avoid Sony cameras for that because of the lossy compression. Long term, I won't be using any DSLR for astro...I'll just be moving to a proper cooled CCD, so it will be a moot point anyway.
I like the actuations, the bigger viewfinder and the Camera in total, but I doubt that my pictures would be improved with it. I learned to master the exposure to the right, I'm not pushing shadows all night long just to prove the abilities of my sensor. I consider it to be my second body, but not my main-tool. The lensrange is a shame against canon and 399 Af Focuspoints? That's marketing, the Canon 70D would have 20 millions on this calculation.
I was kinda standing there in shock for several minutes, half expecting the guy to come back and say he was joking. Never saw him again. It's like we were Crips and Bloods in a tense encounter... in his head.
