Photographic Dynamic Range (PDR) is a More Useful Measure than the DxOMark Landscape Score

Started Apr 17, 2015 | Discussions thread
Detail Man
Detail Man Forum Pro • Posts: 17,073
Re: Photographic Dynamic Range (PDR) is a More Useful Measure than the DxOMark Landscape Score

bclaff wrote:

Detail Man wrote:

bclaff wrote:

Just a couple of point ...

Regarding SNR = 20 see the other clarification.

Regarding E-M5 versus GH2 you intermix two concepts and make it incorrectly appear that the result has to do with FWC when it has entirely to do with read noise only.

In my procedure used in coming up with the information and graphics displayed in this post:

http://www.dpreview.com/forums/post/55668848

Before DxO published their E-M5 test results, any subsequent estimates of E-M5 FWC existed:

I took a small GH2 base-ISO deep shadow image-crop (recorded and posted by Rick'sAstro):

... from an area within this (larger) crop and pushed it in using RAW Therapee 4.x by +5.000 EV.

.

I took a small E-M5 base-ISO deep shadow image-crop (recorded and posted by Rick'sAstro, of the same scene, with the same lighting, using equal sensor-level Exposure):

... from an area within this (larger) crop and pushed it in using RAW Therapee 4.x by +5.345 EV - using a rationale that the E-M5 metering (when used) appeared to select an (approximately) 0.345 EV lower sensor-level Exposure level when presented with the same scene luminance.

My goal was to (roughly) simulate differences that appeared to exist between the two cameras when using the in-camera metering systems in both the E-M5 as well as the DMC-GH2.

The (approximate, "eyeballed only") SNR was estimated by me as being (around) 12 dB (equal to a linear factor of 4) in the above discussed case. I reported what I saw as a result.

Having an (albeit crude) look at GH2 and E-M5 dark-frame RAWs using RawDigger, the Read/Dark noise levels appeared to be fairly close in the value of their Standard Deviations.

.

Later (after DxOMark data was published), Sensorgen reported estimated E-M5 FWC as being 24959 e-, as compared to the DMC-GH2's 11835 e-, and estimated Read Noises of 6.5e-, and 6.0e-, for the E-M5 and the DMC-GH2, respectively. The difference between the (Sensorgen stated) photosite-pitches was a mere 100 nM (pretty close) - normalizing for photosite area would have only a small effect upon the above-stated Read Noise values.

I do not understand the meaning of your statement quoted above:

Regarding E-M5 versus GH2 you intermix two concepts and make it incorrectly appear that the result has to do with FWC when it has entirely to do with read noise only.

(Post DxOMark E-M5 data being published), I noticed that while the Sensorgen estimated Read Noise reported was nearly the same (thus, roughly comparable), the estimated FWCs were around twice as large in the E-M5 (relative to the DMC-GH2), and the (DxOMark published) base-ISO Dynamic Ranges were around 1 EV higher. That is all that I know ...

That's because sensorgen reports read noise in electrons, which came from read noise in DN and gain, and gain came from FWC / (ADC range).
The one stop difference in the Landscape score comes from the one stop difference in engineering dynamic range. Both are 12-bit ADCs and the read noises in DN are about 1.173 and 2.083 respectively.

I apologize for not recognizing any fundamental errors in my thinking (other than, perhaps, my attempts to simulate observed differences of in-camera metering systems by "pushing" the E-M5 deep shadow image-crops by +0.345 EV more). Could you possibly try to explain further ? Your statements above are not "sinking in" (my perhaps somewhat "thick head").

For simplicity I have snipped a lot to write this reply.

First, as they say, "what we have here is a failure to communicate". It probably runs both ways

Well, hopefully we may find some enjoyment and meaning in trying. If not so, that is OK, too.

I have had some interesting "chicken and egg" conversations in the past regarding gain and FWC.

My position is that gain comes first and FWC is a function of gain.

That sems to follow. You spoke of:

... read noise in electrons, which came from read noise in DN and gain, and gain came from FWC / (ADC range).

So, the "DN" and "ADC range" units cancel in a multiplication, leaving some value up to and including the (so-called) "FWC" (in electron charge-units), which is related to the "gain" value.

We can make that moot by sticking to DN and ignoring FWC.

Alright ! Sometimes some unexpected margins affecting upper-most ADU count values - and black-level information (did I read you mention that such was nowadays often "canned"?).

So in DN, which is what we have in the raw data, the E-M5 read noise is 1.07 DN and the GH2 read noise is 2.08 DN for engineering dynamic ranges of 11.9 and 10.9 respectively.

Interesting. I and another person were coming up with Std Devs on the order of 0.8 for the E-M5, and 1.0 for the GH2 (at base ISO on each; 200 and 160, respectively) using RawDigger. Truly ...

This difference in engineering DR carries over to the DxOMark Landmark Scores of 12.3 and 11.3 respectively.

As we increase signal above read noise the higher gain of the E-M5 versus the GH2 would have an effect on the PDR; but in this case the distance is so small no effect is really seen.

It seems like if it were the case (where it comes to the ratio of Read Noises that you report for the GH2 as well as the E-M5, respectively), was reflected with ...

... In this region the subtle interplay of read noise, other noise, small amounts of photon noise, slope due to Full Well Capacity (FWC), etc. is captured ...

... then one (might) expect to see a bit more of an inter-camera effect taking place "there" ?

I suppose it's a good time to point out that I'm not saying the DxOMark figures are always "wrong", particularly in a relative sense.

As with all machinery, it seems not to involve matters so much of "right/wrong" (as applied to subjective psychovisual impressions) - more a healthy awareness of the limitations/errors of certain process(es) that exist - along (perhaps) with some advantageous strong points.

But I am saying it fails when read noise is very low and the conversion gain between two cameras being compared is very different.

Glad that you are onto such things ! It is a valuable and interesting function, and as well a kindness.

Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow