5D3 Photographic Dynamic Range etc.

bobn2 · Mar 10, 2012

surf1 said:
I'll try again.

You got a scene where you need to expose 1/8s at ISO 100 to capture the full range of highlights. The brightest pixel will be 100% saturated = maximum numerical value in the RAW file.

But you need 1/60th because you are shooting hand held. That's 3 stops less light, every stop reduces the amount of light on the pixels by 50%, so your brightest pixels would be only 12,5% saturation.

The amount of light on the sensor is not changing with ISO, but if you increase the ISO 3 stops from 100 to 800, the amplifiers and ADC will be adjusted that a pixel with 12,5% saturation will be the 100% value of the ADC output.

So your highlight pixels with 12,5% saturation will again correspond to the maximum of the 14 Bit output range of the ADC and you can use all those 14 Bits to record the tonal gradation between 0 and 12,5% saturation of the pixels.

Except that you can't. The only sensor that yield enough information to use all those 14 bits ar the Exmor ones. The others could do just fine with a 12 bit or even 10 bit ADC. It is not about gradation in any of these cameras, they are oversupplied with levels, the issue is solely dynamic range, the ratio between the largest signal recordable and the noise.

surf1 said:
Base Noise:
The lower limit of your dynamic range is defined by the noise floor.

If that noise floor were a amplification independent value, then each step ISO increase would reduce the DR by exactly 1 Stop, because it's using only half of the sensor's range.

But in some cameras there is an effect that reduces the base noise with increased amplification and therefore compensates to some extent for the 1 Stop highlight DR. This is shown in this diagram:

Looking at it back to front. It is not that there is a reduction in the base noise with amplification, it is that there is a high level of noise in the ADC and as the amplification is raised it lifts the weak signals above that noise. If the noise were not there, there would be no reason to increase the amplification.

surf1 said:
Look at this diagram and compare the D7000 and the 5DIII:
http://home.comcast.net/~NikonD70/Charts/PDR_Shadow.htm

The D7000 is almost ISO-less, meaning there it gains almost nothing in the shadow range when increasing the ISO.

It gains nothing because it has lost nothing, the ADC noise is low enough that the low level signals are already greater than it.

surf1 said:
The 5D3III on the other hand manages to compensate a lot of the highlight losses up to ISO 3200. It gains 2,5 stops shadow DR while losing 5 stops highlight DR.

It is not 'compensating highlight losses', it is overcoming the huge shadow noise at base ISO. If it didn't have that high level of noise, there would be no 'compensation'.

surf1 said:
But you have to see those values always in combination with the DR at base ISO! Alone, they are deceiving.

If you look at the total DR:
http://home.comcast.net/~NikonD70/Charts/PDR.htm

you see that the 5DIII keeps DR losses minimal between 100 and 800, and only after 3200 ISO it cannot recover any more shadow range and loses 1 stop with every ISO increase.

The D800 starts with much higher ISO, but cannot recover much shadow range, so with increasing ISO the DR of the 2 cameras becomes much closer.

I hope that helped, and the electronic engineers will forgive my rather simplified explanation of the matter.

Just a bit back to front in places.
--
Bob

Ralf Ronander · Mar 10, 2012

I know.

Been there, done that (with my 50D). I just don´t think it´s worth the extra trouble under "normal" circumstances, what you gain is not much in practical terms. IMO. And beyond 3-4 stops most converters can´t cope, and you still get better results upping the ISO.

Ralf Ronander · Mar 10, 2012

I understand what you´re saying.
Guess we have to disagree on what is practically relevant.

Thanks for your answers anyway.

John Sheehy · Mar 10, 2012

bhollis said:
These graphs seem to indicate that the 5D3 has significantly less noise than the D800 between ISO 400 and around ISO 12000. Am I interpreting this correctly?

His numbers for 5D3 read noise are way off. 6400 should be 33+ ADU, for example.

--
John

yabokkie · Mar 10, 2012

sorry I'll check it later ...
if D800 can beat rivals single handed in 1.2x mode (over a half stop advantage).

John Sheehy · Mar 10, 2012

Peter 13 said:
marcus2137 said:

Typically Canons loose about 2/3-1 stop on iso 50, all in the highlight range. Most Canon users typically avoid iso 50. So even if the 5d3 is equal at iso 50 and 100, that's a nice step forward.

Click to expand...

ISO 50 is ISO 100 with changed EXIF - same DR.

Same is true for 5D2.

--
John

bclaff · Mar 10, 2012

John,

I think you may be right about the Canon numbers; but the Nikon look correct to me.

I may have reversed the optical black and clamped data being unfamiliar with Canon.

I'm reworking the numbers.

The read noise have dropped and are posted they are now in line with your value(s).

The PDRs will drop, not posted yet; hopefully no one will "shoot the messenger".

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

RedFox88 · Mar 10, 2012

marcus2137 said:
very interesting because iso 50 is a non-native iso and is supposed to loose DR because of the 'push' the camera does to achieve it. In my light testing is seems that the highlights get cut sooner at iso 50

ISO 50 in RAW is simply ISO 100 over exposed by 1 stop. So you can simulate ISO 50 by simply adding positive exposure bias with ISO 100. If you have a low contrast scene, it is likely you won't over expose the highlights by doubling the exposure time of your ISO 100 exposure. JPG is a different topic and I never shoot JPG let alone use ISO 50 with it.

yabokkie · Mar 10, 2012

my two cents,

probably we can say D800 is better than D4 but it looks
some kind of NR kicked in at about ISO 1600 and 3200 so
how to compare Nikon to Canon we still don't know.

one estimation may be based on accurate measurement at low ISOs,
and extend it "naturally" toward the right.

noirdesir · Mar 10, 2012

Ralf Ronander said:
I know.

Been there, done that (with my 50D). I just don´t think it´s worth the extra trouble under "normal" circumstances, what you gain is not much in practical terms.

I feel that what I gain most is considerable peace of mind of not having to worry about precisely nailing exposure. It helps that a lot of low light situations are inside with somewhat constant lighting (not constant in space but constant in time). With my Nikon D3 (which is 'ISOless' from ISO 800 on), as soon as I hit ISO 800, I know I can switch to manual mode and focus on f-stop and shutterspeed. That thus does not mean that I leave the ISO at 800 all the time, if the highlights are some way away from clipping, I up the ISO, ditto if the metering falls maybe more than 1.5 stops into negative territory. Thus, for most images, I have between zero and 1.5 stops adjustment in the raw converter, occasionally if things change faster than I can adjust the ISO it is up to two stops. And that works good enough in most converters.

It is this knowledge that a stop more or less light will not require my intervention to get a very close to optimal result that is really liberating.

Ralf Ronander said:
And beyond 3-4 stops most converters can´t cope, and you still get better results upping the ISO.

surf1 · Mar 10, 2012

bobn2 said:
bobn2 said:

I hope that helped, and the electronic engineers will forgive my rather simplified explanation of the matter.

Click to expand...

Just a bit back to front in places.
--
Bob

Thanks Bob!
As I said, mechanical engineer

noirdesir · Mar 10, 2012

Jon Rty said:
Then why does the DR of the D800 decrease more rapidly?

As already said, it is the other way around: the DR of the 5D III increases more because the shadow noise decreases for these cameras with increasing ISO and because Canon's do tend to have lower shadow noise (at ISO settings where the AD converter noise becomes small enough).

But from a practical point of view, the DR of the 5D III might be larger than that of the D800 at higher ISO settings. However, for a given mid-tone noise and shadow noise, the D800 can be used at a lower ISO setting than the 5D III for a given exposure. Thus, for this given exposure, the D800 will have the higher DR. But who wins this game of DR for a given mid-tone or shadow noise (for a given exposure) depends on how we weigh the shadow noise vs. the mid-tone noise (ie, it cannot be objectively answered).

ejmartin · Mar 10, 2012

yabokkie said:
actually Bill's read noise and PDR curves show that
the old architecture used in 1DX and D4 is the right one to choose
over Sony because it delivers better performance at ISO 800 and above.

Are the curves displaying pixel dynamic range, or dynamic range at a fixed image scale relative to frame size? The former favors lower pixel count cameras (for a given sensor size), while the latter is appropriate to comparing images from cameras with different pixel counts.

--
emil
--

http://theory.uchicago.edu/~ejm/pix/20d/

bclaff · Mar 10, 2012

Emil,

ejmartin said:
Are the curves displaying pixel dynamic range, or dynamic range at a fixed image scale relative to frame size?

The later, naturally. See the links under Further Reading on the charts for further info (although you know this stuff).

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

Jon Rty · Mar 10, 2012

If I'm not mistaken, read noise of the 5DIII but not the D800 decreases while shot noise increases. And the 5DIII has so much higher read noise at base ISOs to beginn with, which is why it's worse at all ISOs. As the read noise of the D800 doesn't change, the ISO setting doesn't matter. But shot noise increases as you get less and less light, which is why the D800 also increases in noise. The PDR in bills chart at any given ISO is the wedge between clipped highlights and blacks that are beyond the SNR limit of 10.

noirdesir said:
As already said, it is the other way around: the DR of the 5D III increases more because the shadow noise decreases for these cameras with increasing ISO and because Canon's do tend to have lower shadow noise (at ISO settings where the AD converter noise becomes small enough).

But from a practical point of view, the DR of the 5D III might be larger than that of the D800 at higher ISO settings. However, for a given mid-tone noise and shadow noise, the D800 can be used at a lower ISO setting than the 5D III for a given exposure. Thus, for this given exposure, the D800 will have the higher DR. But who wins this game of DR for a given mid-tone or shadow noise (for a given exposure) depends on how we weigh the shadow noise vs. the mid-tone noise (ie, it cannot be objectively answered).

Jon Rty · Mar 10, 2012

Is it not because the read noise of the D800 stays the same, while it decreases for the 5DIII?

bobn2 said:
Because it starts much higher, at 100 ISO.

The way ISO is implemented on most cameras, they will lose a stop of DR with each stop of ISO. Cameras such as the D3s, D4 and 5DIII have so much noise at base ISO that the DR level is already cut to what it would be several stops higher. Thus, it doesn't fall as ISO is raised, since it's already cut.
--
Bob

bobn2 · Mar 11, 2012

Jon Rty said:
Is it not because the read noise of the D800 stays the same, while it decreases for the 5DIII?

Sure, but the 5DIII 'stays the same' after some ISO setting. The 5DIII read noise only falls because it has a burden of noise to carry in the first place.
The point is, it's advantage D800, not the other way round.

And Bill graphs show that the read noise does fall on the D800, but from a very low start. Nikon has used this sensor more like Pentax did in the K-5 than they did in the D7000
--
Bob

Jon Rty · Mar 11, 2012

Doesn't the 5DIII read noise fall due to sensor/signal amplification architecture? I'm very well aware that the D800 has the advantage, just as the D7000 had in APS-C sensors. I'm not trying to claim anything different.

bobn2 said:
Sure, but the 5DIII 'stays the same' after some ISO setting. The 5DIII read noise only falls because it has a burden of noise to carry in the first place.
The point is, it's advantage D800, not the other way round.

And Bill graphs show that the read noise does fall on the D800, but from a very low start. Nikon has used this sensor more like Pentax did in the K-5 than they did in the D7000
--
Bob

Steen Bay · Mar 11, 2012

bclaff said:
bclaff said:

Are the curves displaying pixel dynamic range, or dynamic range at a fixed image scale relative to frame size?

Click to expand...

The later, naturally. See the links under Further Reading on the charts for further info (although you know this stuff).

Regards
--
Bill (visit me at http://home.comcast.net/~NikonD70/ )

Don't quite understand.. your output size is fixed (app. 1mp), but what about your 'noise floor', would that also be fixed (at SNR 20:1) if the output size was e.g. 8mp instead, or does your criteria vary with the output size, so that PDR remains constant? On DxO the criteria is fixed (I think..) at SNR 1:1, both at the native output size (screen) and when normalized/downsampled to 8mp (print), meaning that the DR will increase when the output size is reduced.

Steen Bay · Mar 11, 2012

bobn2 said:
Jon Rty said:

Is it not because the read noise of the D800 stays the same, while it decreases for the 5DIII?

Click to expand...

Sure, but the 5DIII 'stays the same' after some ISO setting. The 5DIII read noise only falls because it has a burden of noise to carry in the first place.
The point is, it's advantage D800, not the other way round.

And Bill graphs show that the read noise does fall on the D800, but from a very low start. Nikon has used this sensor more like Pentax did in the K-5 than they did in the D7000

But a SNR 20:1 noise floor, like used here, shouldn't that make the read noise almost completely irrelevant?

Edit - No, SNR 20:1, that can be achieved with just 400 photons with zero read noise, so e.g. 20e- read noise will of course have an effect.

5D3 Photographic Dynamic Range etc.

Forum Pro

Leading Member

Leading Member

Forum Pro

Forum Enthusiast

Forum Pro

Forum Pro

Forum Pro

Forum Enthusiast

Forum Pro

Well-known member

Forum Pro

Veteran Member

Forum Pro

Senior Member

Senior Member

Forum Pro

Senior Member

Veteran Member

Veteran Member

Similar threads

Keyboard shortcuts