EOS 7D Mark II & Photokina

[sarangiman]

Re: EOS 7D Mark II & Photokina

There's another arena Canon may do better in w/ the 5D3/1Dx AF module. There are 5 high-precision dual-cross-type sensors down the middle. I wonder if these actually have demonstrably higher precision for wide aperture shooting compared to Nikon's central AF points. To my knowledge, no one's tested this rigorously.

I've also noticed some body/lens combos where there'd by systematic back/front-focus based on which direction the focus element was moving. With both Canon and Nikon systems. I wish someone would test this stuff in a systematic manner so we could see where these problems do or do not exist.

Ultimately, though, the inability to track subjects in the X-Y plane have led to many more missed shots with my 5D3 compared to (my limited shooting thus far with) the D810, so that matters to me.

 

[jrista]

Re: EOS 7D Mark II & Photokina

Oh, and jrista:

I meant to respond to your comment at some point about how you shoot at ISO 400 and average images to help with Canon DR. Honestly, at first glance I thought it was quite funny, but I did the math and you were on to something there. I'll see if I can write out the actual math to show where that method benefits, and where it doesn't.

But in the meantime, I just wanted to say:

Some people will say that extra DR doesn't help landscape photographers that much b/c the shot noise in the shadows makes shadows look like higher ISO shots. So you should HDR anyway. Ok, valid point; just that with Canon you have this problem *on top of* extra noise due to read noise.

But I'd argue that even in such situations, the higher DR sensor still helps dramatically. Why? B/c you don't actually have to merge different exposures to get HDR. You can just shoot a bunch of the *same* exposures - since almost everything is more likely to be above the incredibly low noise floor - then average them, then post-process to taste. This avoids the issue of masking/HDR software merging different exposures, where dark/bright boundaries can cause issues in the algorithms as to which image to use pixels from. Of course, you may still have to mask for selective brightening/darkening, but this is much easier - to me anyway - than merging different exposures for HDR.

With Canon & this method, you'd be fighting read noise on top of shot noise, requiring you to average many more images. Or increase the exposure to get some of the darkest subjects in your scene above the noise floor to begin with.

Make sense?

Makes total sense. In the past, I'd thought that the 5D III's read noise was much more random than it is. That was what people were raving about back when it was first released. Now that I own it, it's clear to me that the read noise is still heavily banded, and some of that banding is fixed. The problem with averaging fixed information is it strengthens, where averaging random information effectively bleeds it out. So I totally agree, the clean, random photon shot noise that occurs way down into the deep shadows is far more effective for averaging multiple shots together than with a Canon camera.

I'm still not quite sure of the "ISO-less" nature of the Exmor. I know that's what it's generally called on DPR, but if that was the case, why does Exmor still employ per-pixel amplifiers? There has to be a benefit to amplifying the RAW signal at the pixel, before read, if Exmor is still doing that. Otherwise, the entire thing would be 100% digital, and all ISO settings would be achieved as you say...simply by digitally boosting ISO 100 shots. I agree that shadow detail and color fidelity in a D800 is better, by far as I now see, than a 5D III...however to use it in a true ISO-less manner. You have to be losing color fidelity in the highlights at some point, no?

Regarding the 3D AF...I knew about it's capabilities. I'm curious though. Whenever I've played with Nikon cameras in a store, the AF points always seemed rather tightly clustered in the center of the frame. The point spread seems rather low, in other words. One of the things that is nice about Canon's 61pt AF system is the very wide point spread...I can AF way out well into the midframe, even on a full frame camera. Is that still the case with Nikon? Are their AF points still tightly clustered and small?

There is also the fact that the 61pt AF system has a ton of cross-type points, and the five double-cross type in the center. I don't know how 3D AF would affect bird and wildlife photography, I'd have to give it a try. I do know that having high precision AF points, and having cross type points spread throughout the entire frame, is a huge bonus over what the 7D offered.

 

[sarangiman]

Re: EOS 7D Mark II & Photokina

Makes total sense. In the past, I'd thought that the 5D III's read noise was much more random than it is. That was what people were raving about back when it was first released. Now that I own it, it's clear to me that the read noise is still heavily banded, and some of that banding is fixed. The problem with averaging fixed information is it strengthens, where averaging random information effectively bleeds it out. So I totally agree, the clean, random photon shot noise that occurs way down into the deep shadows is far more effective for averaging multiple shots together than with a Canon camera.

Glad you see that now. But even if it were perfectly random, you'd still have to average many, many more Canon shots to use shadows that you could use from a D800/810 or Sony A7R without any averaging whatsoever. And the nice thing about shot noise in shadows is that, well, our visual system is more tolerant to them b/c they're shadows and not as bright and visible as midtones and highlights. So I'll often put up with the shadows after a huge lift of a D810 file, simply b/c the shot noise is nowhere near as image detail killing as the read noise from my 5D shots.

I'm still not quite sure of the "ISO-less" nature of the Exmor. I know that's what it's generally called on DPR, but if that was the case, why does Exmor still employ per-pixel amplifiers? There has to be a benefit to amplifying the RAW signal at the pixel, before read, if Exmor is still doing that. Otherwise, the entire thing would be 100% digital, and all ISO settings would be achieved as you say...simply by digitally boosting ISO 100 shots. I agree that shadow detail and color fidelity in a D800 is better, by far as I now see, than a 5D III...however to use it in a true ISO-less manner. You have to be losing color fidelity in the highlights at some point, no?

In the highlights? Did you mean shadows?

There could be a number of answers to this question, and it's probably a combination of things. First of all, your hardware really does have to be 'ISO-less'. This means almost zero downstream read noise. We're not there yet, but we're quite close. Second, there's the issue of ADC bit-depth, FWC, and quantization error. There's a subtle interplay between pixel pitch, FWC, and ADC bit-depth that can lead to increased quantization error if you don't do any amplification above base ISO levels. Too much to go into here, but I bet you can think through it.

Point is: there are still some hardware limitations.

But perhaps more importantly, re-hauling the idea of ISO (which is really still a holdover from the days of film) is a big step, and combining that with exposing for the highlights, ETTR, etc. is a very different way of shooting. And camera manufacturers radically innovating and disrupting is not an every-day occurrence.

Furthermore, I wonder how many people at the camera companies are fully aware of just how ISO-less their cameras are. You need to do some sophisticated analysis (photon transfer) to really get at this, but it's possible to rigorously quantitate benefits vs. cost. Now, how many of those that are aware & have done this also understand the photographic implications? And what subset of those can actually convince the rest of the people that need convincing to implement this type of feature in the next camera? And who can do it in a clever-enough way that it appears pretty transparent to the end user who, ideally, should only experience the benefits rather than be forced to understand a new way of shooting...?

To do it absolutely properly, I think there'd have to be some significant changes to the entire way the camera meters and then displays the scene. But I won't go into that here. Suffice it to say there'd have to be a number of algorithmic, as well as UI, tweaks. And how often do we see manufacturers make radical changes? The point is: there are already energy barriers to doing this, and these companies tend to be rather slow to begin with.

But someone should, and someone will. In the meantime, there are those of us that just use the camera in very unintuitive ways to still get the benefits of this type of shooting.

Regarding the 3D AF...I knew about it's capabilities. I'm curious though. Whenever I've played with Nikon cameras in a store, the AF points always seemed rather tightly clustered in the center of the frame. The point spread seems rather low, in other words. One of the things that is nice about Canon's 61pt AF system is the very wide point spread...I can AF way out well into the midframe, even on a full frame camera. Is that still the case with Nikon? Are their AF points still tightly clustered and small?

No. Here's a side-by-side showing the spread of AF points between the Nikon D800 and Canon 5DIII:

The only advantages I see with the Canon AF spread are the 4 extra corner points (upper/lower right, upper/lower left), and the higher density.

There is also the fact that the 61pt AF system has a ton of cross-type points, and the five double-cross type in the center. I don't know how 3D AF would affect bird and wildlife photography, I'd have to give it a try. I do know that having high precision AF points, and having cross type points spread throughout the entire frame, is a huge bonus over what the 7D offered.

Yes, as I mentioned in my previous posts. Actually, the 7D had all cross-type points, just not with the spread of the 5DIII/1Dx. The lack of cross-type points does sometimes cause hunting when focusing on horizontal lines (e.g. eyes/eyebrows) in portrait orientation. You can sometimes combat this by rotating your composition slightly in portrait orientation (or rotate slightly when acquiring AF). It's certainly not ideal, but it can help.

And, remember, the cross-type points were literally the only reason I chose the Canon 5DIII over the D800 years ago (actually I also disliked the grip, the D-pad, the LCD, and the Live View implementation - all of which have been fixed in the D810). So I'm with you there. Recently, armed with a very good sense of my AF hit-rate with micro-adjusted fast primes at f/1.4 with the 5DIII (from shooting with it extensively over the past few years), I decided to give the D810 a fair chance with a 35mm f/1.4 lens. My hit-rate actually went up, and that with a 36MP sensor which has a much tighter tolerance for focus error. Furthermore, it enabled a type of shooting I literally couldn't do with the 5DIII, tracking subjects (usually the eye - I like very close-up, off-center comps with the 35/1.4) in three, not one, dimensions. And I would often use well off-center AF points. Yes in tough lighting situations in portrait orientation the outer points would hunt on horizontal detail - the main reason I dismissed the D800 years ago. But when I weighed that against the hit rate at f/1.4 otherwise, and the tracking, it really put the lack of cross-type AF in perspective.

This is not at all meant to be an excuse for Nikon (and most brands') lack of cross-type AF sensors well off-center. It's difficult to design, if I understand correctly, and I applaud Canon for pioneering this. I really hope Nikon (and Sony in their SLT line, for that matter) follow suit.

Now, when I'd previously tried 3D AF tracking, I don't remember being *this* impressed with it. The speed of the tracking is so fast and steady sometimes that my first reaction was to wonder if the camera was using its accelerometer/sensors to along with information about the focal length of the lens to move the AF point. It sometimes feels *that* responsive. Of course, that was easily disproved by having the stationary camera track a fast moving subject. It's possible that a combination of the improved processing speed, further optimized algorithms, and the 91k sensor (as opposed to the 2k sensor in some other models) combined to make this experience this time so impressive. It's hard to say, though, b/c the D800 and D4s also track very well, and without a more objective, rigorous test... I can't say they're any worse or better than the D810. But they were all better at shifting the AF point to stay on the initial subject than the 1Dx in my tests thus far - where the 1Dx would typically play catch-up or lose the subject over time altogether. That's not to say the 1Dx can't/won't perform just as well or better in other scenarios, but it's demonstrably slower than either the D4s or the D800/810 at sticking to the original subject while constantly recomposing the scene by moving the camera. Not that I would ever even consider the 1Dx b/c of its heft, as well as its utterly inane implementation of EC in M mode w/ Auto ISO. Then again, at least it's the one Canon DSLR to have what most other modern ILCs have... what's ironic is that Canon DSLRs need EC in M mode with Auto ISO the most - simply b/c it's more critical to nail ISO on Canon DSLRs to avoid downstream read noise than it is for, say, an A7 or D800 where you can mistakenly leave the camera at ISO 100 in M mode and then just brighten in post (shooting Raw). In other words, you generally want the ISO amplification on Canon DSLRs to be the highest level that doesn't clip highlights so as to maximize the chances of any image data being well above the noise floor.

But again, I digress

All that said, I really would like to do a controlled study to see if Canon's dual cross-type high precision points actually give better precision (repeatability) under various lighting levels compared to the center points on the D810. As well as comprehensive tests on AF precision with various lenses on these systems. Basically, objectify tests that these days too many people offer subjective opinions on...

 

[sarangiman]

Re: EOS 7D Mark II & Photokina

Also, I wanted to provide a real-world example of when 'ISO-less' shooting can help to retain the base ISO dynamic range of the camera at higher ISOs.

Below is a shot of the setting sun taken on a lake in Cambodia on a rough, rocking motorboat. Taken with a Sony A7R with Canon 70-200mm f/4L (non-IS)* + 1.4x II TC + Metabones Smart Adapter III.

Given the constant motion of the boat & the lack of IS on the lens, and the focal length of 280mm, I dared not use a shutter speed below 1/250s. I wanted an aperture of f/11 to guarantee focus on the boats all the way to the sun. For a proper midtone exposure, this meant I had to use ISO 1600. But instead I used ISO 100, so here's what the shot looked like SOOC:

I then selectively applied what amounted to a 4 EV push in the shadows, and varying levels of exposure boost in others. Importantly, not allowing the sun to blow out. This allowed me to capture a rather high dynamic range scene even at a high ISO setting - albeit only a 'simulated' one, if you will. Meaning I didn't actually use a high ISO, although the focal plane exposure was set to something more along the lines of what the camera would've used at a higher ISO (you're probably beginning to see why the ISO term itself is becoming a source of confusion when shooting in such an odd manner, as I was alluding to in my previous post).

My take home point -- to those who say you don't care about base ISO DR b/c higher ISO DR matters more to you, I say: here's an example where you can approach the base ISO DR of the camera at higher ISOs, albeit only if you shoot in this very strange, seemingly broken manner But there's a bigger point here - the low downstream read noise that enables high base ISO DR also allows for higher DR at all ISOs if you shoot in this 'ISO-less' manner or if, in the future, manufacturers tune their cameras to work in such an 'ISO-less manner'.

In other words, jrista, here's an example where underexposing by 4 EV (by lowering the ISO 4 EV in M mode) was not only a reasonable thing to do, but the *right* thing to do to save the red channels sampling the sun from clipping.

Some fun math to prove my point better:
Loading the RAW file in RawDigger, there's a 9.5 EV difference between the red channel in the sun vs. in some of the shadows of the people on the boat. Now, let's say I'd gone ahead and used ISO 1600 and picked an exposure such that the sun was just under the clipping point. The A7R & 5DIII both have a DR of ~10 EV (using SNR=1 as the lower cutoff) at ISO 1600, so perhaps those 9.5 EV would have, technically, fit into the DR capabilities of both cameras. But that would've placed the shadows at SNR=1. That's unusable to most people. But at ISO 100, the A7R has ~13 EV pixel-level DR. The red channel of the sun is at about 1 stop below clipping in this shot**, which places the red channel of some of those shadows ~10.5 EV below the clipping point of this camera, which should still be comfortably (2.5 EV) above the noise floor given the 13 EV of DR. If I'd done the same with a 5DIII - which has a pixel-level DR of ~11 - some of my shadows would've been 0.5 EV above the noise floor (of SNR=1). Definitely not where you want to be.

The difference in performance only becomes even more marked with the extra DR of the D810 at ISO 64.

To sum it up again:
The camera with high base ISO DR can still give you high DR in situations that'd typically demand use of high ISOs. Whereas the camera with equivalent high ISO DR but poor base ISO DR would not give you this option at all.



*I gave up on the f/4L IS lens b/c 3 copies I tested would, every now and then, render 1/4 to 1/3 of one side of the image soft... the non-IS is much better behaved in this regard. Also I'm not the first to report this: http://diglloyd.com/articles/LensAndCameraIssues/BrandNewBlur-Canon70_200f4L.html.

**So, yes, I could've boosted exposure more at the time of exposure. But, (1) I was already constrained by the shutter speed & aperture I felt appropriate for that situation, and (2) while yes I could've gotten away with ISO 200 here, I'm not a computer hard-wired in to my camera, so the best gauge I had of the ISO to use was the histogram on the camera - which showed me approaching clipping. And, to be honest, sometimes when I don't want to think too much about the technicalities and want to focus on the moment instead, I'll just set shutter speed & aperture on my A7R for (1) artistic intent and constraints, and (2) to avoid clipping, then simply leave ISO at 100 and raise selectively in post. For high DR scenes anyway. Otherwise I'll keep the camera in Auto ISO mode & use negative EC to avoid highlight clipping. The only cost I *sometimes potentially* pay is quantization error for deep deep shadows at ISO 100, and maybe some tiny amount of downstream read noise influence. But even for the former, I think it's irrelevant for most situations, as only signals lower than 16 photoelectrons have their random fluctuations (shot noise, which'd be ±4 photoelectrons here) represented by less than 1 digital unit/increment at ISO 100. I think I've probably lost any reader interest I still had at this point so...

 

[AmselAdans]

Re: EOS 7D Mark II & Photokina

*snap*

Inspiring post, I never thought about this approach.
But isn't there the high risk of color banding?

Assume you have a 10 bit color channel encoding, giving you 1024 shades of each channel (for this example, it doesn't mind whether its 8, 10, 12 or 14 bit..)
The people on board are quite dark in the original shot, so the color information, e.g., of the blue channel may be concentrated in an interval of [0..140]. When then compensating the underexposure, you expand this interval, as the guys now feature lighter shades of blue in their shirts. From some quick Photoshop color grabbing in your picutres, the interval would now approximately be [0..500].
Thus, this will have a posterization effect at a certain degree. You definitively lose color information by this way.

The tradeoff would be interesting between this color posterization but low ISO setting and no posterization but noise from higher ISO settings.

 

[sarangiman]

Re: EOS 7D Mark II & Photokina

Inspiring post, I never thought about this approach.

Thanks, and hey jrista - there you go. A data point for 'how many people have actually thought of & tested the validity of this approach?'

But isn't there the high risk of color banding?

Assume you have a 10 bit color channel encoding, giving you 1024 shades of each channel (for this example, it doesn't mind whether its 8, 10, 12 or 14 bit..)
The people on board are quite dark in the original shot, so the color information, e.g., of the blue channel may be concentrated in an interval of [0..140]. When then compensating the underexposure, you expand this interval, as the guys now feature lighter shades of blue in their shirts. From some quick Photoshop color grabbing in your picutres, the interval would now approximately be [0..500].
Thus, this will have a posterization effect at a certain degree. You definitively lose color information by this way.

The tradeoff would be interesting between this color posterization but low ISO setting and no posterization but noise from higher ISO settings.

First of all, it's a bit of a fallacy to think that noise comes from higher ISO settings. In this case, holding shutter speed & aperture constant, increasing the ISO would have potentially led to *lower*, not higher, noise. The higher noise from higher ISO settings comes from the typically shorter exposures people use at higher ISO settings. But the type of shooting I was doing - all of that goes out the window. You determine the shutter speed/aperture (the focal plane exposure), and then you choose the highest ISO that doesn't clip highlights. My bigger point was that for some cameras, it's not even critical that you choose this 'highest ISO' at all - just keep it near base and adjust exposure in post-processing.

Now, you're worried about expanding the intervals in low tones by raising the exposure in post. But you should be just as worried about raising those tones via ISO amplification in the camera! Because those low, dark tones also would've had very low, 'granular' signals at the level of the sensor - prior to ISO amplification.

The only difference is that the former (raising in post) runs the risk of some loss of information compared to the latter if your ADC bit-depth is not adequate enough at base ISO to represent the noise in the signal accurately. That's why in my second footnote I brought up the topic of quantization error and making sure that the random fluctuation in the signal (resulting mostly, if not almost exclusively, from shot/photon noise in high performance sensors) are accurately represented in the RAW file. You should be fine as long as the standard deviation of any given pixel signal is represented by at least 1 digital unit/increment. This does fall apart (if I did my math right) for signals lower than 16 or so at ISO 100 for the A7R. But, those are very, very low signals. And I'm still not convinced it actually has a significant, discernible effect.

Fundamental to understanding all this is realizing that when you sample a signal of 4 photoelectrons, the best SNR you can get is sqrt(4) = 2. This is b/c the standard deviation of that signal due to statistical/photon/shot noise (http://en.wikipedia.org/wiki/Shot_noise) is equal to the sqrt of the signal, or 2, in this case. This noise can actually provide some dithering effect, so accurately recording tones generating 3, 4, or 5 photoelectrons is less about actually accurately recording those as 3, 4, and 5 and more about making sure you don't drastically undersample shot noise (which has a dithering effect). In other words, instead of worrying about recording increments of 1 in the digital file, you should worry about recording the increments of the fluctuations of that signal. And for a signal of 4, that means accurately representing a fluctuation of 2 (sqrt(4)), not 1. Of course, you'll probably always be better off with a higher bit-depth ADC (barring other associated deleterious effects related to system design); I'm just saying I don't know how much it matters in most shooting scenarios.

In the words of Emil Martinec, who did a beautiful study/treatise on all of this: "Posterization becomes apparent when the quantization step sufficiently exceeds the width of the noise, the random jumps in tonality due to the noise are no longer able to dither the discrete jumps due to quantization."

Read more about this here: http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3.html#bitdepth

Remember that your digital file is trying to represent a quantized signal (based on # of photoelectrons generated during exposure at any given pixel). As long as your amplification and bit-depth are set up so that 1 digital increment in your RAW file represents 1 or less photoelectrons, you should be pretty accurately representing the original signal. When 1 digital increment represents more than 1 photoelectron (which can be the case for base ISO on full-frame sensors), you *could* run into tonality issues, but you often don't because of shot noise that leads to random fluctuations in the signal that amount to sqrt(signal). Here, it becomes more important that you digitally accurately represent the fluctuations in the signal (noise) rather than the quantized steps in the signal itself. This essentially lessens the need for very high bit-depth ADCs, save for in some outlier use-cases.

Apologies in advance if that made absolutely no sense...

 

[sarangiman]

Re: EOS 7D Mark II & Photokina

Let me reiterate:

There might still be *some* advantages even for the most 'ISO-less' cameras today to using, say, ISO 400 amplification vs. ISO 100 for the darkest of signals. Because there's still some finite read noise in the signal chain. However, it may be academic for many.

And for very very high ISO applications, where every fraction of an electron in noise counts, you'll see benefits to hardware-level ISO amplification. Again, b/c of that finite read noise that is still introduced by the signal processing. Or b/c of other things like conversion gain optimizations (e.g. what's suspected to be the case in the A7S). And when you're trying to make an image with 40-50 photoelectrons of less per pixel... every electron of noise saved counts.

 

[jrista]

Re: EOS 7D Mark II & Photokina

Let me reiterate:

There might still be *some* advantages even for the most 'ISO-less' cameras today to using, say, ISO 400 amplification vs. ISO 100 for the darkest of signals. Because there's still some finite read noise in the signal chain. However, it may be academic for many.

And for very very high ISO applications, where every fraction of an electron in noise counts, you'll see benefits to hardware-level ISO amplification. Again, b/c of that finite read noise that is still introduced by the signal processing. Or b/c of other things like conversion gain optimizations (e.g. what's suspected to be the case in the A7S). And when you're trying to make an image with 40-50 photoelectrons of less per pixel... every electron of noise saved counts.

I agree with this. In this case, this is where Canon cameras are currently superior (at least, to most...I think the D810 is starting to change this). At higher ISO, most Exmor cameras still have ~3e- RN, even more. Canon sensors, on the other hand, drop to 2e- and below at the higher ISO settings, dropping to as little as ~1.3e- on some cameras at very high ISO. Those savings in RN are what give Canon's 1D X, 6D (and even to a small degree, the 5D III) the dynamic range/SNR edge at high ISO. The D810 ultimately drops to 1.3e- at it's highest ISO setting, but it still doesn't manage to eek out more DR than the 5D III, and certainly not as much as the 6D and 1D X.

However, with the D810, Nikon seems to have introduced a bias offset, and they now have a roughly linear falloff in read noise compared to the flat RN curve of the D810. I think that further enhanced the value of using higher ISO settings, vs. using the camera "ISO-less", for anything above ISO 400.

As far as ISO-less applications go, I still think that is currently only the realm of low ISO/bright light shooting. You can underexpose to preserve the highlights, and recover the midtones and shadows. That only really works because 3-5e- RN is trivial compared to 50ke- to 80ke- maximum signal, and even 25ke- to 40ke- midtone signal. I don't think quantization noise is really much of an issue either...that would only really cause a bit of posterization in the really deep shadows...but your unlikely to be lifting them enough to actually see that...not unless you have a LOT of shadow detail to recover.

 

[sarangiman]

Re: EOS 7D Mark II & Photokina

I agree with this. In this case, this is where Canon cameras are currently superior (at least, to most...I think the D810 is starting to change this). At higher ISO, most Exmor cameras still have ~3e- RN, even more. Canon sensors, on the other hand, drop to 2e- and below at the higher ISO settings, dropping to as little as ~1.3e- on some cameras at very high ISO. Those savings in RN are what give Canon's 1D X, 6D (and even to a small degree, the 5D III) the dynamic range/SNR edge at high ISO.

No. And I wish these myths weren't propagated with a resilient persistence as readily as they are on forums. First of all, if you want to talk about 'superior', talk about the Sony A7S and its sub 1 electron read noise.

More importantly, I'd treat those read noise numbers on sensorgen a little more softly. Don't get me wrong - I do value sensorgen - but do consider there are no error bars or stdev quoted. Second, it's not just about (upstream) read noise here. You're quoting fractions of electrons differences, yet ignoring QE differences?

The reality is: there are differences, but they're more in-line with generations of tech; beyond that you're splitting hairs at best. Furthermore, simply quoting the min read noise ignores # of pixels, which matters for sensible, normalized comparisons.

Fortunately, we can look at some test shots taken at the same focal plane exposures:

D810 vs 6D vs 1Dx vs D800E
http://tinyurl.com/n3cke3v

Are those differences - at ISO 12,800 no less - really worth talking about? I guess that's up to you, and if it were me I'd rank performance D810 = 1Dx > 6D > D800E*. I'd also remind you that the D810 is performing on par with the 1Dx despite having twice as many pixels! Yes the D800E is performing slightly worse, but at more reasonable ISOs of 3200 and below, it's a wash.

And this whole 'Canon cameras are superior' attitude... it's like brand religion. And it's demonstrably false. Now let's take a look at some other cameras, including the 5DIII which you think is somehow better than Exmor:

D4s vs A7R vs 5DIII vs D610
http://tinyurl.com/kd7vze3

Pretty marginal differences between the 5DIII, the A7R (Exmor), and the D610. Nikon D4s beats all of them, though. And, actually, at higher ISOs the A7R - after removal of the magenta tint in shadows - perhaps even slightly outperforms the 5DIII in normalized comparisons: http://tinyurl.com/k2dam52. At any rate, it certainly doesn't underperform compared to the 5DIII.

So where is this Canon 'superiority' and 5DIII's 'small degree [of an]... edge' at high ISO?

While we're at it, let's look at some APS-C cameras:

Canon 70D vs Nikon D3300 vs Nikon D7100 vs. Canon Rebel T5i
http://tinyurl.com/orncyb8

The Nikon D7100 (Toshiba sensor) outperforms all of those cameras, and the Nikon D3300 performs better than both the Canon cameras despite 1/3 EV less focal plane exposure (all other cameras received the same focal plane exposure).

I could go on, but I think you get my point...

The D810 ultimately drops to 1.3e- at it's highest ISO setting, but it still doesn't manage to eek out more DR than the 5D III, and certainly not as much as the 6D and 1D X.

Why are we still talking about high ISO DR disadvantage for the low (downstream) read noise camera (D810) when I clearly demonstrated in one of my previous posts that you can get demonstrably more DR at high ISO by underexposing & boosting exposure selectively - something you're less able to do with a Canon DSLR?

Furthermore, according to that site-no-one-speaks-of-here-but-who's-data-the-site-you-quoted-numbers-from-derives-its-data-from, there's no difference in normalized DR between the D810 and the 5DIII at high ISOs.

However, with the D810, Nikon seems to have introduced a bias offset, and they now have a roughly linear falloff in read noise compared to the flat RN curve of the D810. I think that further enhanced the value of using higher ISO settings, vs. using the camera "ISO-less", for anything above ISO 400.

Since you brought it up: I'm a little confused by sensorgen's numbers for the D810, as they disagree with DxO's. Sensorgen gives the D810 lower pixel-level DR than the D800/E, whereas DxO clearly shows the opposite.

As for the value of higher ISO vs. shooting 'ISO-less' (to some degree anyway), that depends on what you care about. Comparing ISO 1600 vs ISO 400 + 2EV (same shutter speed/aperture), the tradeoff is between (1) a couple electrons more noise per-pixel, and (2) two whole stops of DR.

I personally, generally, prefer the flexibility of option (2). But now we're veering off the main point of our discussion, which is that the absence of downstream read noise (high base ISO DR) opens up possibilities otherwise impossible to achieve. And this doesn't come at some high ISO performance cost (compared to Canon). It didn't back when the D800 and 5DIII were introduced, and it doesn't now.

And it's a tenuous thread - if not utterly wrong in some cases - to claim that somehow Canon sensors in general are 'superior' to all others with respect to ISO performance. Compared to Exmor or no Exmor.

*The Nikon files have a bit more magenta in blacks at the highest ISOs, though this could possibly be from ACR's 'Shadows Tint' calibration.

 

[jrista]

Re: EOS 7D Mark II & Photokina

I agree with this. In this case, this is where Canon cameras are currently superior (at least, to most...I think the D810 is starting to change this). At higher ISO, most Exmor cameras still have ~3e- RN, even more. Canon sensors, on the other hand, drop to 2e- and below at the higher ISO settings, dropping to as little as ~1.3e- on some cameras at very high ISO. Those savings in RN are what give Canon's 1D X, 6D (and even to a small degree, the 5D III) the dynamic range/SNR edge at high ISO.

No. And I wish these myths weren't propagated with a resilient persistence as readily as they are on forums. First of all, if you want to talk about 'superior', talk about the Sony A7S and its sub 1 electron read noise.

Whoa. Slow down buddy. First off, I hadn't checked the A7s on Sensorgen. I agree, it has vastly superior RN at high ISO. Pretty amazing, actually. That said, I wouldn't call this a "myth"...there is a Q.E. difference at high ISO, however DESPITE that, the higher RN of the D800/D600/etc. reduces their dynamic range a bit at higher ISO.

The D4 and 1D X (and to a degree the 6D), as well as the A7s all enjoy a meaningful improvement to DR at very high ISO settings thanks to their low RN. It was specifically the increase in DR that I think makes these sensors superior at high ISO. I think the benefit of smaller pixels from a resolution standpoint is quickly lost once you start talking about the highly photon shot noise dominated signals at ISO 6400 and up (and maybe even ISO 3200).

This is one of the reasons I think a multi-layered sensor, vs. a bayer array, with the ability to dynamically bin pixels 2x2 and maybe even 3x3 as you increase ISO would be pretty awesome. The sparseness of the signal at really high ISO definitely increases noise, so the value of pixel count starts to diminish rapidly. Bigger pixels become increasingly valuable. That's why I think the 1D X and 5D III (and D4 and A7s) have a nice edge over the D800 and D600...they all have bigger pixels.

More importantly, I'd treat those read noise numbers on sensorgen a little more softly. Don't get me wrong - I do value sensorgen - but do consider there are no error bars or stdev quoted. Second, it's not just about (upstream) read noise here. You're quoting fractions of electrons differences, yet ignoring QE differences?

The reality is: there are differences, but they're more in-line with generations of tech; beyond that you're splitting hairs at best. Furthermore, simply quoting the min read noise ignores # of pixels, which matters for sensible, normalized comparisons.

It may be slitting hairs. The differences are indeed small, not large as they are at low ISO. However some of these sensors DO enjoy as much as a stop or so additional DR at high ISO over the D800, D600, and many of Sony's cameras. As I said, that seems to be changing with the D810 (and apparently the A7s).

As I mentioned, I think the value of small pixels quickly diminishes once were talking about ISO 6400 or 12800 or higher...then I think the read noise levels become critical.

Fortunately, we can look at some test shots taken at the same focal plane exposures:

D810 vs 6D vs 1Dx vs D800E
http://tinyurl.com/n3cke3v

Are those differences - at ISO 12,800 no less - really worth talking about? I guess that's up to you, and if it were me I'd rank performance D810 = 1Dx > 6D > D800E*. I'd also remind you that the D810 is performing on par with the 1Dx despite having twice as many pixels! Yes the D800E is performing slightly worse, but at more reasonable ISOs of 3200 and below, it's a wash.

And this whole 'Canon cameras are superior' attitude... it's like brand religion. And it's demonstrably false. Now let's take a look at some other cameras, including the 5DIII which you think is somehow better than Exmor:

D4s vs A7R vs 5DIII vs D610
http://tinyurl.com/kd7vze3

Pretty marginal differences between the 5DIII, the A7R (Exmor), and the D610. Nikon D4s beats all of them, though. And, actually, at higher ISOs the A7R - after removal of the magenta tint in shadows - perhaps even slightly outperforms the 5DIII in normalized comparisons: http://tinyurl.com/k2dam52. At any rate, it certainly doesn't underperform compared to the 5DIII.

So where is this Canon 'superiority' and 5DIII's 'small degree [of an]... edge' at high ISO?

While we're at it, let's look at some APS-C cameras:

Canon 70D vs Nikon D3300 vs Nikon D7100 vs. Canon Rebel T5i
http://tinyurl.com/orncyb8

The Nikon D7100 (Toshiba sensor) outperforms all of those cameras, and the Nikon D3300 performs better than both the Canon cameras despite 1/3 EV less focal plane exposure (all other cameras received the same focal plane exposure).

I could go on, but I think you get my point...

Sure, I get the point. I've made the point several times myself...sensitivity is primarily a factor of total sensor area, not pixel size. The primary benefit of smaller pixel size is resolving power, not sensitivity. I think all the DPR links above demonstrate that point excellently. All of the APS-C samples have visibly more noise than the FF samples...and, you are right, there are only small visible differences between the FF sensor samples.

To be honest, the only real differences I can see are the differences on underlying tint. Some are redder, some are greener...and as you noted, the A7r seems to have a bit of a magenta tint. An excellent normalized comparison. I am curious, however, how each of those files handles exposure pushing and pulling, and color tweaking, in post. That is ultimately what I personally am most concerned about...in the end, how the images look is really the result of your processing. Does that magenta tint in the A7r clean up well, or is it difficult to clean up? Which ones handle a little bit of shadow recovery better?

There may simply not be an objective answer to those kinds of questions, simply because it's not really tested for. However, to the editing latitude end...your technique, the ISO-less technique, may really be the answer. I would have to experiment. If your underexposing by five stops then lifting by five stops, as though you shot at an ISO setting five stops higher...is there any actual benefit in improved DR? If you are not using the upper five stops that ISO 400 allows, then the primary difference between doing that...and shooting at 12800 is going to be the read noise levels. If it's 3e- at ISO 400 -5EV, and 1.6e- at ISO 12800, so long as your not clipping your highlights...wouldn't shooting at ISO 12800 directly be better? (Maybe I am misunderstanding your ISO-less technique...anyway, this is an honest question...I'd like to know what the real benefit of shooting ISO 400 -5EV and lifting in post is, given the RN differences.)

The D810 ultimately drops to 1.3e- at it's highest ISO setting, but it still doesn't manage to eek out more DR than the 5D III, and certainly not as much as the 6D and 1D X.

Why are we still talking about high ISO DR disadvantage for the low (downstream) read noise camera (D810) when I clearly demonstrated in one of my previous posts that you can get demonstrably more DR at high ISO by underexposing & boosting exposure selectively - something you're less able to do with a Canon DSLR?

Furthermore, according to that site-no-one-speaks-of-here-but-who's-data-the-site-you-quoted-numbers-from-derives-its-data-from, there's no difference in normalized DR between the D810 and the 5DIII at high ISOs.

I think I need to see an example of the ISO-less technique to understand the benefits of -5EV+selective boost, vs. just shooting at an ISO five stops higher.

As for sensorgen...I prefer the data they provide as it's UNcooked DXO data. DXO cooks their numbers when they package them up into their neat little reports. DXO has changed how they cook things behind the scenes at times, and a lot of it is black box...so I don't really trust it. I prefer raw, untainted measurements...which I think sensorgen provides. I agree, they could be more complete. SDtevs and the like would be very useful.

As for the D810 vs. 5D III at high ISO, if you compute the total amount of light gathered by the sensor at ISO 12800, the D810 actually does end up pulling ahead;

7380px*4920px*387e-/px = 14,074,663,680e-
5760px*3840px*526e-/px = 11,634,278,400e-

Per-pixel DR is lower on the D810, so you would clip the signal faster...however, if we applied your ISO-less technique, that really wouldn't be a problem. That's usually my greatest concern at really high ISO...clipping the signal, and having enough DR to reproduce good color fidelity and reduce noise. At these high ISO settings, the differences between cameras may boil down to 1/3rd of a stop...but, I think that 1/3rd of a stop can be and is valuable.

However, with the D810, Nikon seems to have introduced a bias offset, and they now have a roughly linear falloff in read noise compared to the flat RN curve of the D810. I think that further enhanced the value of using higher ISO settings, vs. using the camera "ISO-less", for anything above ISO 400.

Since you brought it up: I'm a little confused by sensorgen's numbers for the D810, as they disagree with DxO's. Sensorgen gives the D810 lower pixel-level DR than the D800/E, whereas DxO clearly shows the opposite.

The problem with DXO is we simply don't know exactly how they derive their chart numbers. That's one of the things I like about Sensorgen...it's just a raw data dump. Granted, that leaves it up to the reader to interpret, but...at least there isn't any special weighting or unknown mathematical formulas applied.

I suspect normalization plays a role in the differences, as I don't think any information on sensorgen is based on normalized results.

As for the value of higher ISO vs. shooting 'ISO-less' (to some degree anyway), that depends on what you care about. Comparing ISO 1600 vs ISO 400 + 2EV (same shutter speed/aperture), the tradeoff is between (1) a couple electrons more noise per-pixel, and (2) two whole stops of DR.

I understand that. However, you have to actually make use of the extra DR. Underexposing at a lower ISO and boosting can avoid highlight clipping, and that is one of the primary concerns about shooting at high ISO. However, if you deal with ETTR properly, or use something like UniWB, the chances of clipping at any ISO are low. Then, the reduction in read noise does become meaningful. If your white point at 400+2EV is the same as at ISO 1600, you lost DR because of the higher RN level at ISO 400 (assuming it is higher...not really the case with the D800/E...definitely the case with the 5D III, D810, A7s, 1DX, etc.) When your saturation point gets low enough, even a fraction of an electron difference in read noise actually becomes quite meaningful as far as dynamic range goes. One electron at very high ISO could mean the difference of a couple stops of DR.

I personally, generally, prefer the flexibility of option (2). But now we're veering off the main point of our discussion, which is that the absence of downstream read noise (high base ISO DR) opens up possibilities otherwise impossible to achieve. And this doesn't come at some high ISO performance cost (compared to Canon). It didn't back when the D800 and 5DIII were introduced, and it doesn't now.

And it's a tenuous thread - if not utterly wrong in some cases - to claim that somehow Canon sensors in general are 'superior' to all others with respect to ISO performance. Compared to Exmor or no Exmor.

*The Nikon files have a bit more magenta in blacks at the highest ISOs, though this could possibly be from ACR's 'Shadows Tint' calibration.

I agree, option 2 is very flexible. I like the idea. I wouldn't call 3-5e- RN the "absence" of read noise...however it is so much lower than 33e- or 38e- read noise that it is quite significant. I still believe that there is value to having <2e- RN at high ISO vs. 3.3e-...the DR formula 20*log(sigSat/RNrms) becomes increasingly sensitive to RNrms the smaller sigSat gets.

The lines of differences in high ISO performance have definitely blurred. The D810 and A7s are performing at a higher level than the D800/E, D600, etc. were at high ISO. Again, I am not sure how the DPR samples demonstrate anything other than differences in color noise tint at high ISO...I'd like to know how the files respond to editing. Not all files respond the same.

 

[Gantz]

Re: EOS 7D Mark II & Photokina

Wow and i thought i have too much time today.

 

[NancyP]

Re: EOS 7D Mark II & Photokina

Speaking of which, the Sony A7S looks like a killer astro-landscape and extreme low-light camera.

In the "Can you top this" mode concerning sensitivity, check out this news article and links:
http://petapixel.com/2014/09/02/researchers-find-way-to-capture-photos-in-almost-pitch-black-at-only-2-photons-per-pixel/

That should read "Researchers find way to capture photos in almost pitch black at only **** 0.2 **** photons per pixel.

 

[Canonicon]

Re: EOS 7D Mark II & Photokina

I will release what we have in our possession after the announcement if it turns out we were wrong.

after the announcement.. well that is pretty useless don´t you think?

i guess you meant after the NDA expires... that would be today.

 

[Loswr]

Re: EOS 7D Mark II & Photokina

I will release what we have in our possession after the announcement if it turns out we were wrong.

after the announcement.. well that is pretty useless don´t you think?

i guess you meant after the NDA expires... that would be today.

I suspect CR Guy meant exactly what he said. It would make more sense if you bothered to quote the complete statement:

If what we posted turns out not to be true, it will be the greatest hoax of specs that I have seen in the 6 years we’ve been around. I will release what we have in our possession after the announcement if it turns out we were wrong.

In other words, if the source of his information turns out to be correct, he protects that source. If the information was false, he releases the 'greatest spec hoax in 6 years' so we can all have a laugh.

Get it now? Or...did you get it before and just selectively quoted his post to be a troll?

 

[Xero]

Re: EOS 7D Mark II & Photokina

What our information doesn’t specify is whether or not it’s the identical sensor to the EOS 70D.

From the other post, "The sensor does have new technology".

How can you stand by what you said and contradict it at the same time?

I see what you're saying, I didn't word that part of the post correctly. I will amend it.

Please don't take this the wrong way... but I won't believe anything until there is an official announcement.

Canon is VERY! good at keeping things secret... remember DPAF? This was the defining feature of the 70D and there was zero warning it was coming.... I wouldn't even be surprised if Canon had some fake specs circulating around to see where any leaks were coming from... The only thing I am sure of is that the rumours will not be 100 percent accurate....

I have to agree with this statement. I feel the specs we all have must to be wrong. The suggestion the DPAF will come on the 7d mkII without a touch screen is not logical. How would you pick focus point for the AF without it. Not to mention the fact that all lens and bodies prices are dropping across the board. Something big is coming and we have no idea what it is.... I hope.

 

[KacperP]

Re: EOS 7D Mark II & Photokina

That's what you do in industrial espionage - feeding with different types of bs each suspected groups/ppl separately and see what info would leak.