[FozzFoster]

Very cool article found at pentaxrumors.com..

Photographic Dynamic Range versus ISO Setting
Pentaxrumors Post

Very interesting to see the jump in dynamic range around ISO 700 for the new Pentax accelerator chipped cameras (add in the K-70, KP, or K-1ii)!

---

Last edited by FozzFoster; 12-10-2018 at 08:53 AM.

[stevebrot]

Photons-to-photos has slowly been fleshing out their Pentax testing with evaluation of the discontinued bodies and it is good to see the 645D, K-3II, and KS-2 added to the list.


Steve

(...appreciate P2P's analysis, but still not sure regarding real world application...)

[scottDee]

Madness. I'm curious how exactly it's being measured. I could have missed it, but this is all I saw:

Data is measured from raw files taken to my specifications and contributed by people from around the world.

Looks like a serious, motivated guy. He deserves many thanks.

[stevebrot]

Data is measured from raw files taken to my specifications and contributed by people from around the world.

Looks like a serious, motivated guy. He deserves many thanks.

Yes, he is definitely serious.

The analysis is both simple and complex. The simple part is that Bill Claff works from the numeric data stored in the RAW files for captures done under standard conditions. The RAW files are user-submitted. The complex part is how he infers NR, scaling, and so on from a set of numbers since the data are essentially opaque with no explicit clues as to what modifications have been applied in the A/D conversion and initial scaling to 12 or 14 bit digital.* For the patient, he includes a primer on the P2P site:

Sensor Analysis Primer

(Actively wondering why a tech-savvy site would continue to use characters that render as � ?)

FWIW, P2P is followed closely by PentaxRumors and associated rumors sites as a source of a steady trickle of rumorable information. They are also widely cited and debated on digital photography forums when subjects such as base ISO come up, though it is often not clear what the charts actually say.**


Steve

* All data are scaled at least to arbitrary clipping values for black and white as part of the A/D conversion. It is the choice of those values that heavily influence capture character.

** My pet peeve is comparing performance at a particular ISO with the inference that the numbers mean the same across brands...they don't.

---

Last edited by stevebrot; 12-10-2018 at 10:20 AM. Reason: punctuation

[FozzFoster]

I'm curious how exactly it's being measured.

Another important consideration that they state:
"Note that the x-axis is ISO Setting and not a "measured" value."

Different camera manufactures use different definitions of ISO.. ISO 100 of one company may be different than the ISO of another company..

[stevebrot]

Different camera manufactures use different definitions of ISO.. ISO 100 of one company may be different than the ISO of another company..

In other words, ISOs are always "apples to carpet color" when doing point-by-point comparison across brand. The take-away is to not do that comparison.


Steve

[FozzFoster]

The take-away is to not do that comparison.

The comparison can still be useful in terms of maximum/minimum dynamic range and rate-of-change of dynamic range throughout the ISO steps.
However, those data can be taken with a grain of salt relative to their starting position to the x-axis.. doesn't mean the rest of the information derived from those data aren't useful.

[beholder3]

Another important consideration that they state:
"Note that the x-axis is ISO Setting and not a "measured" value."

Different camera manufactures use different definitions of ISO.. ISO 100 of one company may be different than the ISO of another company..

More caveats to think about:

Recall that this value is normalized for the Circle OfConfusion (COC) that is appropriate for the sensor size.
PDR is the dynamic range you would expect in an 8x10 print viewed at a distance of about arms length.

1. It is worthy of a whole new thread to start a discussion on what a CoC "appropriate for the sensor size" should be. There are a lots of opinions and no generally accepted facts.
2. 8x10 has a diagonal of 12.8 inches or 32.5 cm. Ignoring the fun of "measurements" that are based on "about arms length" you can still roughly try to interpret it as something that is about 2x the given image diagonal. At half that distance the human eye can resolve about 6 Mpx. That makes his measurements based on about 1,5 MPx perception. Dxo got some flak for using 8 MPx normalizations...

My definition of Photographic Dynamic Range is a lowendpoint with an SNR of 20 when adjusted for the appropriate Circle OfConfusion (COC) for the sensor.

Ignoring all data below SNR 20 instead of 1 is a completely arbitrary choice based on the authors opinion. And yes, this will skew results. Unreliable.

And it has been discussed before:

1. These are theoretical measurements only, which can (like any "rule of thumb" approach) be close to photographic reality, but often they might not. Unreliable.
2. They also ignore completely any sensor weaknesses (think the Nikon Z banding problem) with banding artifacts. For a photographer one or two stops of "dynamic range" might be completely unusable damaged images, but the numbers will still make it appear as "good". Unreliable.

[stevebrot]

More caveats to think about:

Yes...there is the massaging and private definitions and arbitrary stuff too. I don't have a deep enough interest in this stuff to dive deep and sort of leave the validity of the P2P graphs and comparisons in the same bucket as the DXO Mark equivalents. They are what they are and what they are is not obvious or particularly clear, but may make for interesting discussion and reason to post all sorts of photos showing what may or may not be evidence of the validity of the graph or its inferences.


Steve

(...how does one know the "normal analog range" or "signal in electrons"...)

[Igor123]

Sure, these data may or may not reflect practical reality, but its still fun to sort the list by iso or EV and to see the pentax "entry level aps-c" K70 before ANY canon camera at all (including FFs), just one nikon (the DS) and the K1 II on 5th place beaten only by MFs. (Yeah yeah I know it's the accelerator chip shenanigans but still =) )

(Personally I'd like it better if those placements were for the DR instead, but I guess we can't cheat physics, especially with the same sensors as the others, but either way I'm super happy with the DR and ISO of my K1 I so don't take me wrong =) )

[TonyW]

Sure, these data may or may not reflect practical reality, but its still fun to sort the list by iso or EV and to see the pentax "entry level aps-c" K70 before ANY canon camera at all (including FFs), just one nikon (the DS) and the K1 II on 5th place beaten only by MFs. (Yeah yeah I know it's the accelerator chip shenanigans but still =) )

(Personally I'd like it better if those placements were for the DR instead, but I guess we can't cheat physics, especially with the same sensors as the others, but either way I'm super happy with the DR and ISO of my K1 I so don't take me wrong =) )

I believe the data, at least Bill Claff's do actually reflect reality as they are a measure of real world data with a reasonable degree of accuracy of what you would expect a particular camera system to capture in terms of DR

Dxomark on the other hand have a different method of measuring DR (no less valid but...) I believe on photographing a backlit Stouffer test object resulting in a higher than expected DR level

For example Dxomark rate the DR of the Pentax K1 @ ISO 100 as 14.6 Ev. So they are getting higher DR values than would be expected for a 14 bit capture system. Does this mean anything significant in the real world, perhaps not. EIther the testing methods just work out to be a little OTT or could the camera system be actually capturing more than 14 bits?

Bill Claff’s testing methods suggests a photographic DR of about 11.43 Ev which sounds somewhere near plausible to me. I have tested my own 645z against his measurements using a Sekonic meter and found that we are very close (his probably more accurate than mine). This kind of breeds confidence in me that he is on the same page as I am (in fact probably a chapter or two in front!). Cannot compare Dxomark for the 645z as they pulled it and made some excuses (from memory -3Ev in real world vs Dxomark) - it did come out at the time as the best they had seen so go figure!

TBH I do not think that it really matters too much as long as you are comparing apples with apples by using one set of test data to estimate differences. So use either but do not try and mix and match.

In the real world there are a myriad of reasons to choose a particular camera and buy into a system. Most systems will have some benefits that others envy and vice versa. I doubt that anyone is really going to swap a system for a two stop DR advantage alone

[stevebrot]

For example Dxomark rate the DR of the Pentax K1 @ ISO 100 as 14.6 Ev. So they are getting higher DR values than would be expected for a 14 bit capture system. Does this mean anything significant in the real world, perhaps not. EIther the testing methods just work out to be a little OTT or could the camera system be actually capturing more than 14 bits?

The larger question would be how more than 14 EV are represented when only 14 bits are available to write to. That is a hard limit unless DXO Mark is doing some sort of additive range between color channels or they are depending on the RAW processor to gently call something other than what it is (which they do) in a way that enhances the S/N ratio. That is just one of the disconnects in regards to extrapolating sensor behavior from RAW data analysis. The truth is that there is no way to say with any certainty what the camera did before those numbers are written to file or what was flushed from the sensor to the A/D convertor.


Steve

[stevebrot]

Bill Claff’s testing methods suggests...

I am curious whether anyone reading this thread has actually corresponded with Bill Claff with intent to provide sample files? I am curious about the capture methodology and optical requirements. After all, low contrast or poor resolution degrades apparent DR.


Steve

[TonyW]

I am curious whether anyone reading this thread has actually corresponded with Bill Claff with intent to provide sample files? I am curious about the capture methodology and optical requirements. After all, low contrast or poor resolution degrades apparent DR.

Steve

for various reasons including him already producing 645z figures did not think worth doing. I know that the did ask around the forums and got some rude comment back from here (from memory that is and hope to be wrong!)so I guess he moved on

[TonyW]

The larger question would be how more than 14 EV are represented when only 14 bits are available to write to. That is a hard limit unless DXO Mark is doing some sort of additive range between color channels or they are depending on the RAW processor to gently call something other than what it is (which they do) in a way that enhances the S/N ratio. That is just one of the disconnects in regards to extrapolating sensor behavior from RAW data analysis. The truth is that there is no way to say with any certainty what the camera did before those numbers are written to file or what was flushed from the sensor to the A/D convertor.


Steve

Absolutely agree we really do not know what is happening (do they fully (really and truly ) disclose their testing methodology?), but we can only hope and trust that their testing is consistent and repeatable time after time camera after camera.