Sony's New a7RII Camera Delivers World's First Back-Illuminated FF Sensor

[emko]

The amount of misinformation that is bandied around here regarding the Sony sensor is quite staggering. I've just returned from a four day trip to North Wales, shooting in Snowdonia and took a camera fitted with the Mighty Exmor alongside my usual Canon.

For a start the whole "14 stoops of DR" compared to 12 is so misleading. The Exmor sensor clips to white at exactly the same point as the Canon on a like for like exposure. In fact if we are going to be anal about it the Exmor clips a tiny bit earlier. So much for more dynamic range. The best way to think about the two sensors is that the DR is basically identical but you have more shadow latitude in the Exmor. The Canon will fall apart in the shadows before the Exmor, but whether or not you consider the lack of tonality that comes from the big lift acceptable is up to you.

The Exmor evangelists don't talk about the fragility of the highlights do they ? Or how it loses blue saturation faster than the Canon. And f your answer is to under expose to make the highlights more robust, that's your decision; it's not necessarily how I want to work.

Here are two 100% crops from the Fairy Glen in Snowdonia. Which one is from the mighty Exmor ? It didn't have a cat in hell's chance of holding the highlights coming from the water never mind how much I under exposed.

Of course due to the noise that 'can be seen even in the bokeh or unlifted shadows' you will be able to tell straight away which is the Canon. These shots are the same exposure and converted with no adjustments applied.

My conclusion at present is that if you are working within the acceptable range of the sensor the Canon has the better "IQ" for landscape photography, and I'm someone who is after the new Pentax FF camera when it finally arrives. I hope it doesn't have the 36mp Exmor in it.

umm? one has over exposed area and the other does not but it also has a area that is soo? it looks like the light source changed or was blocked.

 

[jrista]

I did not mean either jrista to seem like you aren't objectively correct. Obviously you are. My first post was simply meant to show my annoyance at the overzealous reactions of how bad Canon is when a new product is released. I don't think you personally are one of them. I think you understand the tech well enough to objectively appreciate that tech. So my post is not aimed at you.

8)

 

[kphoto99]

Here are two 100% crops from the Fairy Glen in Snowdonia. Which one is from the mighty Exmor ? It didn't have a cat in hell's chance of holding the highlights coming from the water never mind how much I under exposed.

My guess is that the top one is Canon. Also a much better lens on the first one.

 

[raptor3x]

This puts a limit on your overall tonality. With high read noise, your tonality is diminished...across the board. The 1D X, with 38.5e- RN at ISO 100, has a maximum tonality (differentiable tones) of 2340 (90101/38.5). That is barely more than 11 stops, or 2^11 (which is 2048). The A7s, on the other hand, has a maximum tonality of 7103 (155557/21.9). That is closer to 13 stops (2^13 = 8192). The increase in differentiable tones in the A7s is what people like me find valuable. That improvement does not just exist in the shadows. It exists throughout the entire signal. It exists in the highlights as much as it exists in the shadows...more importantly, it exists in the midtones, where I think it is actually most valuable.

You're confusing dynamic range with tonal range. They're related, but tonal range (how many levels of grey can be represented) needs to take into account the quantization of noise into discrete levels. The difference in tonal range is nowhere near as large as you're suggesting.

 

[jrista]

This puts a limit on your overall tonality. With high read noise, your tonality is diminished...across the board. The 1D X, with 38.5e- RN at ISO 100, has a maximum tonality (differentiable tones) of 2340 (90101/38.5). That is barely more than 11 stops, or 2^11 (which is 2048). The A7s, on the other hand, has a maximum tonality of 7103 (155557/21.9). That is closer to 13 stops (2^13 = 8192). The increase in differentiable tones in the A7s is what people like me find valuable. That improvement does not just exist in the shadows. It exists throughout the entire signal. It exists in the highlights as much as it exists in the shadows...more importantly, it exists in the midtones, where I think it is actually most valuable.

You're confusing dynamic range with tonal range. They're related, but tonal range (how many levels of grey can be represented) needs to take into account the quantization of noise into discrete levels. The difference in tonal range is nowhere near as large as you're suggesting.

I'm talking about the maximum number of tones that are allowed by the dynamic range of the camera. In an actual signal, you also have photon shot noise, and I'd say the photon shot noise in the signal is going to affect the actual tonal counts in any given image much more than quantization noise. Dynamic range is the space within which the signal fits on the hardware...but it isn't the same AS the signal.

Actual image signals are going to be more similar, for sure...but that wasn't my point. Sporgon implied that Sony cameras have problems with highlights. Because of the increased dynamic range, that is totally false. You can easily shift highlights down by a third of a stop, a half stop, a full stop if you prefer, drop those highlights squarely in the linear response range of the sensor, and lift the shadows a stop, half stop, or third stop to compensate...and poof, any potential issue with highlights in a Sony camera (or Nikon or any other camera that uses an Exmor) is gone. A one stop lift with Exmor shadow data is a no brainer...and even in the actual image signal with all of it's photon shot noise, there is plenty of tonality.

Also, regarding the quantization of the signal. Current consumer grade cameras use 14-bit ADC units. That allows a tonal range of 2^14, which is above the maximum potential allowed by the read noise levels of any of these cameras.

 

[meywd]

This puts a limit on your overall tonality. With high read noise, your tonality is diminished...across the board. The 1D X, with 38.5e- RN at ISO 100, has a maximum tonality (differentiable tones) of 2340 (90101/38.5). That is barely more than 11 stops, or 2^11 (which is 2048). The A7s, on the other hand, has a maximum tonality of 7103 (155557/21.9). That is closer to 13 stops (2^13 = 8192). The increase in differentiable tones in the A7s is what people like me find valuable. That improvement does not just exist in the shadows. It exists throughout the entire signal. It exists in the highlights as much as it exists in the shadows...more importantly, it exists in the midtones, where I think it is actually most valuable.

You're confusing dynamic range with tonal range. They're related, but tonal range (how many levels of grey can be represented) needs to take into account the quantization of noise into discrete levels. The difference in tonal range is nowhere near as large as you're suggesting.

I'm talking about the maximum number of tones that are allowed by the dynamic range of the camera. In an actual signal, you also have photon shot noise, and I'd say the photon shot noise in the signal is going to affect the actual tonal counts in any given image much more than quantization noise. Dynamic range is the space within which the signal fits on the hardware...but it isn't the same AS the signal.

Actual image signals are going to be more similar, for sure...but that wasn't my point. Sporgon implied that Sony cameras have problems with highlights. Because of the increased dynamic range, that is totally false. You can easily shift highlights down by a third of a stop, a half stop, a full stop if you prefer, drop those highlights squarely in the linear response range of the sensor, and lift the shadows a stop, half stop, or third stop to compensate...and poof, any potential issue with highlights in a Sony camera (or Nikon or any other camera that uses an Exmor) is gone. A one stop lift with Exmor shadow data is a no brainer...and even in the actual image signal with all of it's photon shot noise, there is plenty of tonality.

I am way over my head here, but I think what Sporgon meant with the highlight issue is that they don't actually have the DR they claim to have, I understand what you say, that if you can lift the shadows then its because you have more DR in the image, but you also said that they have better DR because they have less noise (or at least that's an explanation of how they did it)

The dynamic range of a camera CANNOT be decoupled from it's noise. Dynamic range is a reference to noise: MaximumSignal/CameraReadNoise. <-- That IS dynamic range, right there. Canon cameras have more read noise. Therefor, the dynamic range of a Canon is plain and simply NOT the same as a Sony.

So maybe they have less DR, but have lower read noise, which means the shadows are in a better condition, but the highlights are sacrificed, now I am not disputing the end result, that you can lift shadows, or that the image produced contains - at least virtually - a higher DR.

 

[jrista]

This puts a limit on your overall tonality. With high read noise, your tonality is diminished...across the board. The 1D X, with 38.5e- RN at ISO 100, has a maximum tonality (differentiable tones) of 2340 (90101/38.5). That is barely more than 11 stops, or 2^11 (which is 2048). The A7s, on the other hand, has a maximum tonality of 7103 (155557/21.9). That is closer to 13 stops (2^13 = 8192). The increase in differentiable tones in the A7s is what people like me find valuable. That improvement does not just exist in the shadows. It exists throughout the entire signal. It exists in the highlights as much as it exists in the shadows...more importantly, it exists in the midtones, where I think it is actually most valuable.

You're confusing dynamic range with tonal range. They're related, but tonal range (how many levels of grey can be represented) needs to take into account the quantization of noise into discrete levels. The difference in tonal range is nowhere near as large as you're suggesting.

I'm talking about the maximum number of tones that are allowed by the dynamic range of the camera. In an actual signal, you also have photon shot noise, and I'd say the photon shot noise in the signal is going to affect the actual tonal counts in any given image much more than quantization noise. Dynamic range is the space within which the signal fits on the hardware...but it isn't the same AS the signal.

Actual image signals are going to be more similar, for sure...but that wasn't my point. Sporgon implied that Sony cameras have problems with highlights. Because of the increased dynamic range, that is totally false. You can easily shift highlights down by a third of a stop, a half stop, a full stop if you prefer, drop those highlights squarely in the linear response range of the sensor, and lift the shadows a stop, half stop, or third stop to compensate...and poof, any potential issue with highlights in a Sony camera (or Nikon or any other camera that uses an Exmor) is gone. A one stop lift with Exmor shadow data is a no brainer...and even in the actual image signal with all of it's photon shot noise, there is plenty of tonality.

I am way over my head here, but I think what Sporgon meant with the highlight issue is that they don't actually have the DR they claim to have, I understand what you say, that if you can lift the shadows then its because you have more DR in the image, but you also said that they have better DR because they have less noise (or at least that's an explanation of how they did it)

The dynamic range of a camera CANNOT be decoupled from it's noise. Dynamic range is a reference to noise: MaximumSignal/CameraReadNoise. <-- That IS dynamic range, right there. Canon cameras have more read noise. Therefor, the dynamic range of a Canon is plain and simply NOT the same as a Sony.

So maybe they have less DR, but have lower read noise, which means the shadows are in a better condition, but the highlights are sacrificed, now I am not disputing the end result, that you can lift shadows, or that the image produced contains - at least virtually - a higher DR.

I'm not sure what that means: "- at least virtually - a higher DR."

Exmor has more dynamic range. Dynamic range is a hardware thing. It's a matter of the maximum charge capacity of each pixel in the sensor, vs. the minimum amount of noise the system produces. If you do that thing everyone hates, stick the camera body cap on the camera, and snap a dark frame, you will have a recording of how much noise the system has. Pop that body cap off, and take a number of successive brightly exposed images until you find the exposure where your highlights just start to clip, and you have found the maximum charge capacity.

Now, before that bright image is read out, there is a quantity of electric charge held in each pixel. That charge is your full well capacity. Divide that charge by the average amount of charge in each pixel from your dark frame...and that is dynamic range. That's all it is. It's not a complicated concept. It's a hardware concept, but it's not complicated.

Now, lets say you take a photo of an 18% gray card. You properly illuminate the card, set your camera to AWB, use ISO 100, fill the frame with it, and expose. The signal in each pixel (the charge in each pixel, the electron count) is going to fall somewhere in the middle of the dynamic range of the camera...about half-way between the read noise level and the point at which you found the signal started to clip. The SIGNAL of the image has a signal to noise ratio. The image is a signal...technically speaking, it does not have dynamic range...it just has an SNR. Like every signal, that signal has it's own noise, on top of the read noise of the system. The SNR is determined by dividing the strength of the signal, by the square root of itself....plus any additional noise in the system. This is an important distinction between dynamic range, and SNR. SNR's are a signal, and the signal itself has noise.

Dynamic range is a hardware trait...its the space within which you can move that signal around. You can expose longer, increase the strength of the signal. That shifts the amount of charge in each pixel towards the clipping point. You can expose shorter, reduce the strength of that signal. That shifts the amount of charge in each pixel toward the read noise. When your read noise is high, you run into it sooner than when it is low. When it is low...you have more room within which to shift the signal around.

If two cameras have the same charge capacity per pixel...let's say 100,000e-. One has 40e- at ISO 100, the other has 4e- at ISO 100. Now, with both cameras, when we take a number of bright exposures to find where the signal clips...they will both clip at the same point. The exposures for barely clipped images should be identical. The cameras differ because of the difference in read noise. The camera with 40e- has 11.32 stops of dynamic range. The camera with 4e- has 14.65 stops of dynamic range.

Alright, here is a real-world example. The Sony A7s has a maximum signal strength at ISO 100 of 155557e-, and read noise of 21.9e-. The 1D X has a maximum signal strength at ISO 100 of 90101e-, and read noise of 38.5. Without even doing the math, it should be obvious that the A7s has more room for signal, plain and simple. A "highlight" pixel in the A7s is going to have more charge than a highlight pixel in the 1D X. The lowest amount of charge in a shadow pixel is going to be lower in the A7s than the lowest amount of charge in a shadow pixel in the 1D X. The A7s has more dynamic range. Plain and simple. There is more room there to move the signal around than in the 1D X.

There is a discrepancy in pixel size here, but we could normalize that. With the same size pixels, the FWC of the A7s would be 104961-, and the relative read noise would be 14.8e-. At this point, the A7s would have pixels the same size as the 1D X, and the same image size. Again, the A7s has more dynamic range. That is obvious from the still larger maximum charge and the still lower read noise. More room to move your signal around in before you actually push the shutter button. You need to preserve more highlight detail? No problem. You have the room to do that, and still have shadows just as good as you can get with the Canon.

Exmor does not sacrifice highlights to improve the shadows. That's not what dynamic range is. Dynamic range is a range...it's a space. It's the container within which the signal of an image fits, and room within which to move that signal around, and make it do what you want it to do.

 

[meywd]

This puts a limit on your overall tonality. With high read noise, your tonality is diminished...across the board. The 1D X, with 38.5e- RN at ISO 100, has a maximum tonality (differentiable tones) of 2340 (90101/38.5). That is barely more than 11 stops, or 2^11 (which is 2048). The A7s, on the other hand, has a maximum tonality of 7103 (155557/21.9). That is closer to 13 stops (2^13 = 8192). The increase in differentiable tones in the A7s is what people like me find valuable. That improvement does not just exist in the shadows. It exists throughout the entire signal. It exists in the highlights as much as it exists in the shadows...more importantly, it exists in the midtones, where I think it is actually most valuable.

You're confusing dynamic range with tonal range. They're related, but tonal range (how many levels of grey can be represented) needs to take into account the quantization of noise into discrete levels. The difference in tonal range is nowhere near as large as you're suggesting.

I'm talking about the maximum number of tones that are allowed by the dynamic range of the camera. In an actual signal, you also have photon shot noise, and I'd say the photon shot noise in the signal is going to affect the actual tonal counts in any given image much more than quantization noise. Dynamic range is the space within which the signal fits on the hardware...but it isn't the same AS the signal.

Actual image signals are going to be more similar, for sure...but that wasn't my point. Sporgon implied that Sony cameras have problems with highlights. Because of the increased dynamic range, that is totally false. You can easily shift highlights down by a third of a stop, a half stop, a full stop if you prefer, drop those highlights squarely in the linear response range of the sensor, and lift the shadows a stop, half stop, or third stop to compensate...and poof, any potential issue with highlights in a Sony camera (or Nikon or any other camera that uses an Exmor) is gone. A one stop lift with Exmor shadow data is a no brainer...and even in the actual image signal with all of it's photon shot noise, there is plenty of tonality.

I am way over my head here, but I think what Sporgon meant with the highlight issue is that they don't actually have the DR they claim to have, I understand what you say, that if you can lift the shadows then its because you have more DR in the image, but you also said that they have better DR because they have less noise (or at least that's an explanation of how they did it)

The dynamic range of a camera CANNOT be decoupled from it's noise. Dynamic range is a reference to noise: MaximumSignal/CameraReadNoise. <-- That IS dynamic range, right there. Canon cameras have more read noise. Therefor, the dynamic range of a Canon is plain and simply NOT the same as a Sony.

So maybe they have less DR, but have lower read noise, which means the shadows are in a better condition, but the highlights are sacrificed, now I am not disputing the end result, that you can lift shadows, or that the image produced contains - at least virtually - a higher DR.

I'm not sure what that means: "- at least virtually - a higher DR."

Exmor has more dynamic range. Dynamic range is a hardware thing. It's a matter of the maximum charge capacity of each pixel in the sensor, vs. the minimum amount of noise the system produces. If you do that thing everyone hates, stick the camera body cap on the camera, and snap a dark frame, you will have a recording of how much noise the system has. Pop that body cap off, and take a number of successive brightly exposed images until you find the exposure where your highlights just start to clip, and you have found the maximum charge capacity.

Now, before that bright image is read out, there is a quantity of electric charge held in each pixel. That charge is your full well capacity. Divide that charge by the average amount of charge in each pixel from your dark frame...and that is dynamic range. That's all it is. It's not a complicated concept. It's a hardware concept, but it's not complicated.

Now, lets say you take a photo of an 18% gray card. You properly illuminate the card, set your camera to AWB, use ISO 100, fill the frame with it, and expose. The signal in each pixel (the charge in each pixel, the electron count) is going to fall somewhere in the middle of the dynamic range of the camera...about half-way between the read noise level and the point at which you found the signal started to clip. The SIGNAL of the image has a signal to noise ratio. The image is a signal...technically speaking, it does not have dynamic range...it just has an SNR. Like every signal, that signal has it's own noise, on top of the read noise of the system. The SNR is determined by dividing the strength of the signal, by the square root of itself....plus any additional noise in the system. This is an important distinction between dynamic range, and SNR. SNR's are a signal, and the signal itself has noise.

Dynamic range is a hardware trait...its the space within which you can move that signal around. You can expose longer, increase the strength of the signal. That shifts the amount of charge in each pixel towards the clipping point. You can expose shorter, reduce the strength of that signal. That shifts the amount of charge in each pixel toward the read noise. When your read noise is high, you run into it sooner than when it is low. When it is low...you have more room within which to shift the signal around.

If two cameras have the same charge capacity per pixel...let's say 100,000e-. One has 40e- at ISO 100, the other has 4e- at ISO 100. Now, with both cameras, when we take a number of bright exposures to find where the signal clips...they will both clip at the same point. The exposures for barely clipped images should be identical. The cameras differ because of the difference in read noise. The camera with 40e- has 11.32 stops of dynamic range. The camera with 4e- has 14.65 stops of dynamic range.

Alright, here is a real-world example. The Sony A7s has a maximum signal strength at ISO 100 of 155557e-, and read noise of 21.9e-. The 1D X has a maximum signal strength at ISO 100 of 90101e-, and read noise of 38.5. Without even doing the math, it should be obvious that the A7s has more room for signal, plain and simple. A "highlight" pixel in the A7s is going to have more charge than a highlight pixel in the 1D X. The lowest amount of charge in a shadow pixel is going to be lower in the A7s than the lowest amount of charge in a shadow pixel in the 1D X. The A7s has more dynamic range. Plain and simple. There is more room there to move the signal around than in the 1D X.

There is a discrepancy in pixel size here, but we could normalize that. With the same size pixels, the FWC of the A7s would be 104961-, and the relative read noise would be 14.8e-. At this point, the A7s would have pixels the same size as the 1D X, and the same image size. Again, the A7s has more dynamic range. That is obvious from the still larger maximum charge and the still lower read noise. More room to move your signal around in before you actually push the shutter button. You need to preserve more highlight detail? No problem. You have the room to do that, and still have shadows just as good as you can get with the Canon.

Exmor does not sacrifice highlights to improve the shadows. That's not what dynamic range is. Dynamic range is a range...it's a space. It's the container within which the signal of an image fits, and room within which to move that signal around, and make it do what you want it to do.

By virtually I mean that the read noise is lower, now both sensors record the same data but the lower read noise allows it be more usable, so the DR of capturing the signal is the same, but the total hardware\system DR is lower in the Canon due to the higher read noise.

now I don't know which camera/sensor he used, but may the one he used had a lower maximum signal strength than the Canon, and a lower read noise.

 

[jrista]

By virtually I mean that the read noise is lower, now both sensors record the same data but the lower read noise allows it be more usable, so the DR of capturing the signal is the same, but the total hardware\system DR is lower in the Canon due to the higher read noise.

Ah. Sorry...it seemed like you were saying the DR of the Sony was lower...I didn't understand that.

now I don't know which camera/sensor he used, but may the one he used had a lower maximum signal strength than the Canon, and a lower read noise.

Yeah...not sure exactly how the differences in FWC are going to affect tonality. I mean, from a charge perspective, a camera that can accumulate more charge is going to have a stronger signal...and that is a good thing. That's a really tough thing to nail down. Actual final image tonality is going to depend on each image. There are...well, effectively infinite images that can be made with any given camera.

I was speaking more to the limits of tonality.

In a camera with a 12-bit ADC, the ultimate limit is going to be 2^12. In a camera with a 14-bit ADC, the ultimate limit is going to be 2^14, and 2^16 for a 16-bit ADC. Quantization noise plays a role when it comes to digitizing the signal...there will always be some amount of quantization noise, but how much diminishes as the difference between the bit depth of the ADC, and the effective bit depth of the signal, increases (i.e. when a quantization error occurs, if your effective signal bit depth is 11, and your ADC bit depth is 16, the quantization error is a tiny fraction of any given step in tonality as limited by noise).

Actual tonality of an image is going to have to account for photon shot noise. That is going to differ from image to image, and even differ from one area of an image to another. There isn't any nailing that down...you can't describe the capabilities of a camera that way.

Dynamic range is how we describe the capabilities in a meaningful, comparable way. More DR is more DR...and since it's the space within which you can fit the signal, more DR means you can protect highlights more if you need to, and not lose on the other end. A lot of people seem to think it's only about the shadows. Actually, it's about both the shadows and the highlights...always has been. We just talk about shadow lifting, because once we have avoided clipping...well, that's where the rest of the recoverable signal is...in the shadows.

 

[meywd]

By virtually I mean that the read noise is lower, now both sensors record the same data but the lower read noise allows it be more usable, so the DR of capturing the signal is the same, but the total hardware\system DR is lower in the Canon due to the higher read noise.

Ah. Sorry...it seemed like you were saying the DR of the Sony was lower...I didn't understand that.

now I don't know which camera/sensor he used, but may the one he used had a lower maximum signal strength than the Canon, and a lower read noise.

Yeah...not sure exactly how the differences in FWC are going to affect tonality. I mean, from a charge perspective, a camera that can accumulate more charge is going to have a stronger signal...and that is a good thing. That's a really tough thing to nail down. Actual final image tonality is going to depend on each image. There are...well, effectively infinite images that can be made with any given camera.

I was speaking more to the limits of tonality.

In a camera with a 12-bit ADC, the ultimate limit is going to be 2^12. In a camera with a 14-bit ADC, the ultimate limit is going to be 2^14, and 2^16 for a 16-bit ADC. Quantization noise plays a role when it comes to digitizing the signal...there will always be some amount of quantization noise, but how much diminishes as the difference between the bit depth of the ADC, and the effective bit depth of the signal, increases (i.e. when a quantization error occurs, if your effective signal bit depth is 11, and your ADC bit depth is 16, the quantization error is a tiny fraction of any given step in tonality as limited by noise).

Actual tonality of an image is going to have to account for photon shot noise. That is going to differ from image to image, and even differ from one area of an image to another. There isn't any nailing that down...you can't describe the capabilities of a camera that way.

Dynamic range is how we describe the capabilities in a meaningful, comparable way. More DR is more DR...and since it's the space within which you can fit the signal, more DR means you can protect highlights more if you need to, and not lose on the other end. A lot of people seem to think it's only about the shadows. Actually, it's about both the shadows and the highlights...always has been. We just talk about shadow lifting, because once we have avoided clipping...well, that's where the rest of the recoverable signal is...in the shadows.

Also maybe the problem with the highlight is optical and not related to the sensor, or in how the sensor is collecting color data, anyway back to your comparison of the 1D X and the a7S, you say the a7S has a read noise of 21.9e- and when you equate the pixel size the read noise becomes 14.8e-, wouldn't the read noise be higher with a smaller pixel?

 

[jrista]

Also maybe the problem with the highlight is optical and not related to the sensor, or in how the sensor is collecting color data, anyway back to your comparison of the 1D X and the a7S, you say the a7S has a read noise of 21.9e- and when you equate the pixel size the read noise becomes 14.8e-, wouldn't the read noise be higher with a smaller pixel?

Well, how much read noise there is really depends on how the system works. It could be higher, it could be lower, that is largely determined by ADC and other electronics, operating frequency, etc. I was just trying to compare on a normalized area basis...so if you took an area of an A7s pixel that was the same as a 1D X pixel, without assuming anything else about the readout system, the amount of noise added during readout of the full pixel relative to the smaller area would be 14.8e-. In Canon's cameras, smaller pixels actually seem to be associated with less read noise. Even in the presence of higher readout frequency, which can increase read noise (i.e. the 7D II has 12.9e- RN and a higher readout frequency than the 7D with 8.3e- RN...both are still considerably less than the readout noise of a 5D III or 1D X). So, I think the comparison is fair, and the relative reduction in RN for comparison purposes is valid given the context.

As for highlights, every sensor has a natural response curve. There is often a small non-linear range at the bottom, where the read noise is (so it usually doesn't matter)...there is usually a very linear range, there is the falloff range, and there is the "rebound" range. The rebound range occurs in some designs, usually CCDs, due to how continued photon incidence and the excess charge from them is bled off. The signal will hit it's maximum, then rebound a bit back down as excess charge is removed by antiblooming technology or something similar. The actual dynamic range of a given camera is usually from the zero charge point to somewhere below the point where the signal starts to go non-linear, and definitely below the lowest rebound point.

Depending on exactly how a manufacturer decides to use the literal signal range of their sensor, and where to cut off the white point, depends on the camera. I have never found any details like that...at least, not for commercial ILC cameras. For the most part, in testing, these cameras generally seem to put the white point somewhere near the top of the linear range. It may be that the A7r does have a little bit of non-linearity in the signal as it grows towards the white point. I wouldn't be surprised, actually...there have been patents filed about purposely doing that, and actually purposely attenuating the highlight range even, as it allows the hihglights to "roll off", rather than just hard clip. I honestly don't know if Sony is doing that. Personally, I have not seen such issues when testing myself. At least, nothing particularly more egregious than what I've seen with my Canon cameras. Every camera, when you get right up near the clipping point, is going to start breaking down. It's rare that all three channels clip at the same time...so you will usually see a loss in color fidelity as one channel out of the three starts to clip, then two. That is usually where you start to see false color or just simple gray in specular highlights...it's just due to the unnatural incongruent clipping of the channels as they run into the white point. I would say it's pretty normal, and I'd expect every RGB camera do encounter some point in the highlights where that occurs.

The way I use these cameras, I find myself running into highlight problems a lot more with my 5D III. I trained myself to ETTR pretty heavily with Canon bodies and preserve as much shadow detail as possible, and now it's just habit. I often get into trouble with my whites and highlights as they all bunch up right near the clipping point, and often one or two channels will indeed clip. Every time I photograph a bufflehead or goldeneye, I tend to clip the highlights in the first couple sets of frames, before I finally remember to tame my ETTR. With Alpha cameras, I started out with the A7r trying to preserve the highlights...and I just don't ETTR as much. It simply isn't as necessary. I have never felt the need, so I haven't run into problems with unnatural highlight tonality.

 

[Marsu42]

There is no such thing as more DR in the shadows, or more shadow latitude. There is simply more DR, more editing latitude. That's it.

Technically true, but not necessarily accurate in the real world b/c my observation is that raw converters are "tuned" for a certain type of histogram distribution, at least ACR is.

That's why for example my 60d has more highlight "latitude" that can be "recovered" from raw files, but it doesn't show up by default but is used for a highlight rolloff that has the unwanted side-effect of earlier clipping. Same with shadows, at least using ML's dual_iso (I don't own an exmor cam): You really need to aggressively "develop" the raw files to expand the "shadows" from the left edge of the histogram with tone curves.

Sony users don't worry about highlights because they don't have to ETTR so heavily to preserve as much signal as possible, like Canon users do.

There's another benefit to this, as I'm not tired to state even though it is contradictory to "a good photog always exposes properly" mantra: More dr lets you expose more lazily w/o the fear of clipping. When not using dual_iso in midday, I often have to take two shots b/c at least my 6d metering is so dodgy it doesn't ettr reliably with +ec.

I freely admit it: I don't find the fun in photography in trying to expose "properly" and be proud of myself if the histogram is filled left to right. I'd rather have some safety space left and concentrate on other matters, esp. when shooting wildlife. Yes, bash away, "8 stops of dr is enough for a capable photog" crowd

The dynamic range of a camera CANNOT be decoupled from it's noise.

True again, but at least Canon has achieved a "nicer" noise pattern with recent cams that responds better to nr algorithsm ... so the dr-noise reletionship has to be viewed after complete postprocessing as this is how our eye and non-tech brain sees the data.

Btw thanks for all your explanations jrista, it's really remarkable how many people bash "more dr" by really weird theories and you're really working hard to contradict them :->

 

[jd7]

I have noticed a pattern on this forum that I think is enlightening. Most of the people who don't care much about dynamic range, or mirrorless vs. DSLR or whatever the latest gripe happens to be seem to skew heavily toward people who actually earn a living in photography. (With Neuro being an exception)

On the other hand, it seems like it's mostly hobbyists with huge disposable income who fit into the category of people obsessed with having the latest, greatest and newest technology and get all worked up when Canon doesn't fit their exact desires.

You should make that observation on different forums. I think you would find that the discrepancy is not nearly as skewed as it is here on CN if you include a much more diverse population of photographers. ...

See my corrected quote above. I was referencing this forum only.

I am not trying to cast aspersions, just simply noting that it seems like most of the regular contributors to this forum whom I respect for actually practicing photography professionally seem must less interested in the esoteric arguments that fuel much of the discussion.

I can't help but feel there is some truth in what Unfocused is saying here - and that it may apply even beyond the boundaries of CR.

I feel like the "Canon is doomed" club approach cameras as consumer tech gadgets - I suspect they have been accustomed to upgrading their computer and their mobile phone every year or two, and their camera too in the "early days" of digital photography, as new tech coming out every year or two really did make a significant practical difference to the gadget. If they haven't had something new for a year or two, their instinct is to be bored and/or sure it must be obsolete. New tech is good simply because it's new tech. The thing is, I don't feel like that sort of dramatic improvement in practical usefulness has been occurring with camera/sensor tech over the last few years, or is happening now. I read references to Canon being boring/uninteresting, but what I don't see is photos taken by non-Canon cameras (well, 35mm format anyway) which make me say wow, that is obviously not taken with a Canon camera. And at the end of the day, it's the photographs produced rather than the tech used, surely?? A camera is a just a tool for photography, and ultimately I do not care if the sensor in my camera used on-die ADC or BSI or whatever, and I don't fundamentally care whether it has a mirror (although as I prefer an OVF to an EVF - at least at this point in the evolution of EVFs - in a sense perhaps I do), etc. Really, I just want a camera which allows me to make photographs I like, and which I enjoy shooting with. (And yes, of course, what makes a camera enjoyable for me to shoot with isn't necessarily what makes a camera enjoyable for anyone else.)

I am sure someone will say - but the IQ from an Exmor sensor is just better!! Well, it has more DR, at least at low ISO and perhaps more recently at high ISO too. That's good (although it's not the only factor in determining IQ). But even if we start with the premise the IQ from an Exmor is inherently "better", look around at photos taken with Canon equipment. Perhaps start by looking around CR, at the shots taken by EML58 and Eldar and others in the 1DX thread, at Sporgon's landscapes, or Pookie's portraits, or the shots posted by Macgyver and many others. If Canon cameras are really so far behind - in terms of the practical results they produce - why don't I see lots of photos around the place which put all of those to shame? Even if Exmor IQ is better, it doesn't seem to regularly translate to substantially better IQ than what the competition provides. Another case in point - the Samsung BSI sensor in the NX1 seems to be good, but I haven't heard of everyone in the APS-C camera segment deserting other manufacturers despite their "old tech" non-BSI sensors.

I do see a benefit in having higher DR than what Canon currently offers, but unlike the "Canon is doomed" club I don't see the current competition offering such substantial practical benefit to my photography that I feel desperate to move away from Canon gear.

To be clear:

I am always interested to know about new tech - for curiosity's sake if nothing else. All else being equal I am happy to take any technical advances on offer. I appreciate the lengths that some people have gone to in many posts to actually demonstrate what some of the tech out there can do, eg jrista's thread some time back with RAWs taken with an Exmor sensor to show the latitude to lift shadows, rather than just make bald and extreme assertions. I'm sure the day will come when I have a camera which allows me that sort of latitude to lift shadows, and I'm sure I will not be unhappy to have it!

I am sure there are people out there for whom the difference in DR really does make a difference to the photograph they do. I'm sure jrista is right to suggest some pros (as well as non-pros) have changed to SoNikon sensors for that reason. And if the increased DR is going to provide a practical benefit for your photography, go for it if you can!

I am sure there are people who will argue that even if you can achieve comparable results with a Canon sensor by use of filters / reflectors / lighting / multiple exposures and blending / etc, using a SoNikon just makes it easier and that's valuable to them. I can accept there are probably situations in which that is correct, but I'm not sure how often those situations arise - at least for me (YMMV). And anyway, ultimately all photography equipment has limits and trade-offs, so it's a case of understanding the limits and trade-offs inherent in your equipment, and knowing techniques to work around them as best you can.

I look forward to seeing what the A7R II can do, what others may offer, and the next round of Canon cameras. I will be interested to see what IQ advances are being brought to the table by the different manufacturers.

Hhhmmm, that turned into more of a rant than I'd planned. And perhaps when we get to see images from the a7R II I'll end up deciding it does offer a substantial practical photographic benefit. I will be interested to see!

 

[meywd]

Thanks for all the explanation jrista, I think I understood that,

 

[unfocused]

I have noticed a pattern that I think is enlightening. Most of the people who don't care much about dynamic range, or mirrorless vs. DSLR or whatever the latest gripe happens to be seem to skew heavily toward people who actually earn a living in photography. (With Neuro being an exception)

I think you would find different distributions of professional's opinions on that subject depending on what it is they shoot and how they shoot. I think many professional landscape photographers have overwhelmingly chosen other brands over the last few years.

Indeed, and that's why there is a whole science about getting data on people's opinions, and it doesn't involve reading some posts on Internet forums that have an inherent heavy bias. Esp. the "pro" vs. "amateur" comparison doesn't compute:

* Imho you just might be able to discern a pattern amongst enthusiasts with deep pockets and time on their hand to experiment around with gear to improve their results (that is probably limited by ability and/or experience). But on the other hand, there are enough users around shooting with a 50/1.8 happily ever after.

* But for "pros" ... how many full-time pros spend their spare time on a general Internet forum, speculating about [CR1] gear that will never arrive? That's instead of spending an evening out with their friends and family or simply enjoying a good glass of wine and a book instead of spending yet more time in front of a screen? If they see the need to switch gear to improve their results, they'll probably just do so w/o feeling the need to explain themselves to anyone.

Yes, I wasn't clear. I was only talking about this particular forum and the participants on it. I waste enough time on this forum. I don't care to read or get involved in others, so I can only speak about this one.

I've been following this forum for many years and have a pretty good idea which contributors earn a living at photography and which don't.

My observation was that most of the contributors who do earn a living from photography seem to be less concerned and get less worked up about the topics that seem to spark such intense feelings among a small minority of forum participants. I've also seen incredibly talented professional photographers driven off this forum by childish, trollish remarks from self-appointed experts. In fact, I consider that one of the most unfortunate things that occurs on this forum -- that professionals who get interested in participating and have much to contribute get fed up and bow out because they can't cut through the noise of obvious idiots.

Honestly, I tend to agree as well with your suggestion about individuals with deep pockets and time on their hands.

I suppose it is my own bias showing, but because I have actually seen the work of contributors like Sporgon and read the advice/recommendations of people like "Private" (and at times have entered into vigorous debates with same) I have a tendency to lend greater weight to their opinions than I do to certain other contributors.

In some respects, I guess it comes down to whether you consider photography to be a science or an art. I know it is both, but I find the "art" side much more interesting.

 

[LetTheRightLensIn]

So maybe they have less DR, but have lower read noise, which means the shadows are in a better condition, but the highlights are sacrificed, now I am not disputing the end result, that you can lift shadows, or that the image produced contains - at least virtually - a higher DR.

No, he has you confused. The highlights are not sacrificed. These are digital sensors and they are linear. The difference in DR comes about all from read noise. The Exmor (and some others now too) have less late stage read noise and thus more DR at low to mid-ISO.

 

[LetTheRightLensIn]

By virtually I mean that the read noise is lower, now both sensors record the same data but the lower read noise allows it be more usable, so the DR of capturing the signal is the same, but the total hardware\system DR is lower in the Canon due to the higher read noise.

now I don't know which camera/sensor he used, but may the one he used had a lower maximum signal strength than the Canon, and a lower read noise.

Well that is sort of true. The initial capture stage of the Canon sensor is at least as good so it has the DR there, but by the time the signal is in a form the camera can process it's been damaged. But since it's always damaged and the Canon sensor doesn't contain the part on the sensor itself that could send it out so it doesn't have to be damaged later on makes the fact that the base sensor detector is at least as good.

But I don't think it makes sense to call it virtual DR. It's simply the DR it delivers.

 

[LetTheRightLensIn]

I have noticed a pattern on this forum that I think is enlightening. Most of the people who don't care much about dynamic range, or mirrorless vs. DSLR or whatever the latest gripe happens to be seem to skew heavily toward people who actually earn a living in photography. (With Neuro being an exception)

On the other hand, it seems like it's mostly hobbyists with huge disposable income who fit into the category of people obsessed with having the latest, greatest and newest technology and get all worked up when Canon doesn't fit their exact desires.

You should make that observation on different forums. I think you would find that the discrepancy is not nearly as skewed as it is here on CN if you include a much more diverse population of photographers. ...

See my corrected quote above. I was referencing this forum only.

I am not trying to cast aspersions, just simply noting that it seems like most of the regular contributors to this forum whom I respect for actually practicing photography professionally seem must less interested in the esoteric arguments that fuel much of the discussion.

I can't help but feel there is some truth in what Unfocused is saying here - and that it may apply even beyond the boundaries of CR.

I feel like the "Canon is doomed" club approach cameras as consumer tech gadgets - I suspect they have been accustomed to upgrading their computer and their mobile phone every year or two, and their camera too in the "early days" of digital photography, as new tech coming out every year or two really did make a significant practical difference to the gadget. If they haven't had something new for a year or two, their instinct is to be bored and/or sure it must be obsolete. New tech is good simply because it's new tech. The thing is, I don't feel like that sort of dramatic improvement in practical usefulness has been occurring with camera/sensor tech over the last few years, or is happening now. I read references to Canon being boring/uninteresting, but what I don't see is photos taken by non-Canon cameras (well, 35mm format anyway) which make me say wow, that is obviously not taken with a Canon camera. And at the end of the day, it's the photographs produced rather than the tech used, surely?? A camera is a just a tool for photography, and ultimately I do not care if the sensor in my camera used on-die ADC or BSI or whatever, and I don't fundamentally care whether it has a mirror (although as I prefer an OVF to an EVF - at least at this point in the evolution of EVFs - in a sense perhaps I do), etc. Really, I just want a camera which allows me to make photographs I like, and which I enjoy shooting with. (And yes, of course, what makes a camera enjoyable for me to shoot with isn't necessarily what makes a camera enjoyable for anyone else.)

I am sure someone will say - but the IQ from an Exmor sensor is just better!! Well, it has more DR, at least at low ISO and perhaps more recently at high ISO too. That's good (although it's not the only factor in determining IQ). But even if we start with the premise the IQ from an Exmor is inherently "better", look around at photos taken with Canon equipment. Perhaps start by looking around CR, at the shots taken by EML58 and Eldar and others in the 1DX thread, at Sporgon's landscapes, or Pookie's portraits, or the shots posted by Macgyver and many others. If Canon cameras are really so far behind - in terms of the practical results they produce - why don't I see lots of photos around the place which put all of those to shame? Even if Exmor IQ is better, it doesn't seem to regularly translate to substantially better IQ than what the competition provides. Another case in point - the Samsung BSI sensor in the NX1 seems to be good, but I haven't heard of everyone in the APS-C camera segment deserting other manufacturers despite their "old tech" non-BSI sensors.

I do see a benefit in having higher DR than what Canon currently offers, but unlike the "Canon is doomed" club I don't see the current competition offering such substantial practical benefit to my photography that I feel desperate to move away from Canon gear.

To be clear:

I am always interested to know about new tech - for curiosity's sake if nothing else. All else being equal I am happy to take any technical advances on offer. I appreciate the lengths that some people have gone to in many posts to actually demonstrate what some of the tech out there can do, eg jrista's thread some time back with RAWs taken with an Exmor sensor to show the latitude to lift shadows, rather than just make bald and extreme assertions. I'm sure the day will come when I have a camera which allows me that sort of latitude to lift shadows, and I'm sure I will not be unhappy to have it!

I am sure there are people out there for whom the difference in DR really does make a difference to the photograph they do. I'm sure jrista is right to suggest some pros (as well as non-pros) have changed to SoNikon sensors for that reason. And if the increased DR is going to provide a practical benefit for your photography, go for it if you can!

I am sure there are people who will argue that even if you can achieve comparable results with a Canon sensor by use of filters / reflectors / lighting / multiple exposures and blending / etc, using a SoNikon just makes it easier and that's valuable to them. I can accept there are probably situations in which that is correct, but I'm not sure how often those situations arise - at least for me (YMMV). And anyway, ultimately all photography equipment has limits and trade-offs, so it's a case of understanding the limits and trade-offs inherent in your equipment, and knowing techniques to work around them as best you can.

I look forward to seeing what the A7R II can do, what others may offer, and the next round of Canon cameras. I will be interested to see what IQ advances are being brought to the table by the different manufacturers.

Hhhmmm, that turned into more of a rant than I'd planned. And perhaps when we get to see images from the a7R II I'll end up deciding it does offer a substantial practical photographic benefit. I will be interested to see!

I think it's because they see things that could help things they do more than it's just specs and latest thing excitement (although I'm sure there is some of that too for some).

 

[LetTheRightLensIn]

I have noticed a pattern that I think is enlightening. Most of the people who don't care much about dynamic range, or mirrorless vs. DSLR or whatever the latest gripe happens to be seem to skew heavily toward people who actually earn a living in photography. (With Neuro being an exception)

I think you would find different distributions of professional's opinions on that subject depending on what it is they shoot and how they shoot. I think many professional landscape photographers have overwhelmingly chosen other brands over the last few years.

Indeed, and that's why there is a whole science about getting data on people's opinions, and it doesn't involve reading some posts on Internet forums that have an inherent heavy bias. Esp. the "pro" vs. "amateur" comparison doesn't compute:

* Imho you just might be able to discern a pattern amongst enthusiasts with deep pockets and time on their hand to experiment around with gear to improve their results (that is probably limited by ability and/or experience). But on the other hand, there are enough users around shooting with a 50/1.8 happily ever after.

* But for "pros" ... how many full-time pros spend their spare time on a general Internet forum, speculating about [CR1] gear that will never arrive? That's instead of spending an evening out with their friends and family or simply enjoying a good glass of wine and a book instead of spending yet more time in front of a screen? If they see the need to switch gear to improve their results, they'll probably just do so w/o feeling the need to explain themselves to anyone.

Yes, I wasn't clear. I was only talking about this particular forum and the participants on it. I waste enough time on this forum. I don't care to read or get involved in others, so I can only speak about this one.

I've been following this forum for many years and have a pretty good idea which contributors earn a living at photography and which don't.

My observation was that most of the contributors who do earn a living from photography seem to be less concerned and get less worked up about the topics that seem to spark such intense feelings among a small minority of forum participants. I've also seen incredibly talented professional photographers driven off this forum by childish, trollish remarks from self-appointed experts. In fact, I consider that one of the most unfortunate things that occurs on this forum -- that professionals who get interested in participating and have much to contribute get fed up and bow out because they can't cut through the noise of obvious idiots.

Honestly, I tend to agree as well with your suggestion about individuals with deep pockets and time on their hands.

I suppose it is my own bias showing, but because I have actually seen the work of contributors like Sporgon and read the advice/recommendations of people like "Private" (and at times have entered into vigorous debates with same) I have a tendency to lend greater weight to their opinions than I do to certain other contributors.

In some respects, I guess it comes down to whether you consider photography to be a science or an art. I know it is both, but I find the "art" side much more interesting.

I haven't found the defend all things Canon side to have portfolios any better than the other side. I've even seem some of the ones going on about how the 'DRoners' need to step out of lab, learn how to shoot, etc. end having a portfolio with like 20 shots of OOF and weirdly exposed shots of cats in their backyard and someone they were ragging on, some pretty fine shots taken around the world in all sorts of scenarios. Whatever the case, it's all kind of irrelevant anyway. It'd be like going into a photo discussion forum and saying that you don't give someone's landscape shots any weight if they can't tell you this or that technical info.


And pros do bring stuff up. On the sidelines I'd hear them bring up AF or this or that now and then.

It is true that some are a bit jaded and it's all a job for them, the pay isn't always high and they are fine with just using whatever gets a result that keeps them in the job. For some things like newsprint, just a little bit of image quality can go quite a long way. Some of them are not all that technical and don't even know as much about what this or that HW can do than many amateurs in some regards. So sometimes the excited amateur is more apt to care about certain things, especially once you get past AF and speed. (Although I should point out I once shot next to a Getty shooter who was getting so disgusted with Nikon sensors, this was obviously years ago, that he was very seriously considering switching sides and he was asking all sorts of questions about Canon sensors and more.)

 

[jrista]

There is no such thing as more DR in the shadows, or more shadow latitude. There is simply more DR, more editing latitude. That's it.

Technically true, but not necessarily accurate in the real world b/c my observation is that raw converters are "tuned" for a certain type of histogram distribution, at least ACR is.

That's why for example my 60d has more highlight "latitude" that can be "recovered" from raw files, but it doesn't show up by default but is used for a highlight rolloff that has the unwanted side-effect of earlier clipping. Same with shadows, at least using ML's dual_iso (I don't own an exmor cam): You really need to aggressively "develop" the raw files to expand the "shadows" from the left edge of the histogram with tone curves.

What you are seeing here has nothing to do with dynamic range, and everything to do with gamma and picture style curves applied during demosaicing. This is a "camera profile" consequence...not a factor of dynamic range. True, it does change from raw editor to raw editor...but that is because you are working with a non-linear digital signal.

Open any raw image in a linear editor, like PixInsight, and you might gain a new understanding.

Sony users don't worry about highlights because they don't have to ETTR so heavily to preserve as much signal as possible, like Canon users do.

There's another benefit to this, as I'm not tired to state even though it is contradictory to "a good photog always exposes properly" mantra: More dr lets you expose more lazily w/o the fear of clipping. When not using dual_iso in midday, I often have to take two shots b/c at least my 6d metering is so dodgy it doesn't ettr reliably with +ec.

I freely admit it: I don't find the fun in photography in trying to expose "properly" and be proud of myself if the histogram is filled left to right. I'd rather have some safety space left and concentrate on other matters, esp. when shooting wildlife. Yes, bash away, "8 stops of dr is enough for a capable photog" crowd

My histograms do not always fill every column from left to right. I wouldn't call such a histogram a "proper" histogram, either. Again dynamic range isn't about proper exposure. It's about capability...the ability to have the power to capture difficult scenes when you need to.

The dynamic range of a camera CANNOT be decoupled from it's noise.

True again, but at least Canon has achieved a "nicer" noise pattern with recent cams that responds better to nr algorithsm ... so the dr-noise reletionship has to be viewed after complete postprocessing as this is how our eye and non-tech brain sees the data.

Btw thanks for all your explanations jrista, it's really remarkable how many people bash "more dr" by really weird theories and you're really working hard to contradict them :->

I agree, Canon has made strides in their noise characteristic, or maybe we could call it noise quality. I also believe, having used a variety of demosaicing algorithms now, that not all banding is actually baked into the signal itself...some of it is derived or added due to how the data is demosaiced. Maybe someday Adobe will refactor their deosaicing algorithm and that problem will go away.