Discussion of 'Equivalence'

[Bjørn Rørslett]

In fact, *any* definition used at the first step has to be independent of the end output - otherwise a feedback (or circularity) is present. This is notwithstanding the obvious logical flaws of requiring a *standard size output* (most photographers would never think this way anyway),, ignoring the problem of empty magnification, the underutilisation of any "dynamic range" (and pixels)  in the first step for the output, the impact of light conditions including scene contrast at the time of capture, and a raft of other potential pitfalls. For example, if one tries to capture a perfectly blank wall, no amount of "dynamic range" or pixels or format, large or small, will suffice simply because a photographic image can not be made in the first place as there is nothing for the system to focus on. Contrasts have to be present (and no, this is not related to AF failure, it is basic photography theory). There thus will be a single value of whatever statistic one implements as a measure of "dynamic range" as at it is not determinable it can be set at any value one wishes as it cannot be changed later. It is a singularity.

[David H. Hartman]

* I have seen plenty of interior architectural photography with large dynamic range printed on paper, which has a much lower dynamic range. It can be done.

Yes, it can be done and it does not prove one should or should not use flash.

For well over a century the toe of the film compressed the shadows and the toe of the printing paper compressed the highlights. What was done with film and paper is now done in the camera or in software. The means of capturing and processing the image has changed. The way it's presented on the paper has not.

B&W printing paper has what, a reflective density of 2.0 or about 6.6 EV? With classic Tri-X, N-1 developing and some dodging and burning I could stuff 13 stops on to a sheet of B&W printing paper and make it look good. The EV range of the scene will most frequently exceed the EV range of a reflective print. That's no problem. What's a problem is when the EV range of the scene exceeds the capacity of the capture process.

DXO says my D800 has a range of 14.4 EV. How do I represent that on a sheet of paper with a range of say 7 EV? Levels and curves? LCH (lightness, chroma, hue)? I prefer LCH because LCH separates the contrast from the saturation. Using the Master Lightness I'll start by drawing down the 1/4 tone and raising up the 3/4 tone to create a soft "S" curve. I don't have to clip the shadows or the highlights to have pleasing mid tone contrast. If the full 14.4 EV were compressed evenly to fit the paper the resulting image would be "Calk and Soot."

Dave Hartman

In the film age did the dynamic range of Tri-X and Paper make a difference compared to that of Velvia? Knowing the capacity of one's capture process is just as important today as it was yesterday.

[simsurace]

No, it proves you should always carry a flash.  Of course it is your fantasy and you can set the conditions however you like so using a flash is impossible?  So it proves that if you are photographing interiors with windows and you can't use a flash you should choose an overcast day.  Then you will modify the conditions again - the interior is in Dubai so overcast days are rare? - so as to preserve the conclusion.  The point is not that you can't modify the conditions endlessly, but that doing so is circular: the only reason to accept the conditions is that they are pre-conditions for the conclusion to be true.

That's the thing about examples; they tend to be crafted in order to illustrate an idea.
I merely wanted to give you an example of a situation where the output dynamic range is lower than what the camera records, but still the additional detail that was recorded can be displayed on the output medium. You wish to make this into a universal statement that it was not intended to be. It was an example, and I already conceded to the obvious fact that lighting the dark parts is another way to get around the problem, as is shooting at night ect.

It is a much harder case, in my opinion, to argue in favour of the statement you seem to make; namely that the additional dynamic range is never useful. Still you seem to be insisting on this point, or you try to put forward a set of rules how things should be done, e.g. solving the above dynamic range problem with flash, and excluding other equally valid possibilities.

[Bjørn Rørslett]

"That's the thing about examples; they tend to be crafted in order to illustrate an idea. "

Or see whether a theory can deal with it.

[simsurace]

"That's the thing about examples; they tend to be crafted in order to illustrate an idea. "

Or see whether a theory can deal with it.

We started by dissecting the concept of PDR or similar normalized measures of DR.
Now we got so far as also questioning the concept of DR in general. At least this is what Les is doing, if I understand him correctly. It is not clear to what end? Even if we all agreed on this board that it is meaningless to talk about DR of cameras, this would nevertheless not stop anyone from being interested in it.
I do not understand what the end goal is. To conclude that all theoretical concept are useless?

To summarize, the current discussion revolves around what are useful notions of DR (dynamic range).
- Some people (including you, if I understood your statements correctly) argue that per-pixel DR is useful, but not normalized measures since they lead to confusion. You prefer to think of it in terms of secondary magnification.
- Les Olson argues (again, if I understood him, if not, please correct me and provide clarification) that all DR measures are useless
- My point of view is that depending on the question, either per-pixel DR or PDR can be useful
- Jack has argued that while PDR is useful, it should not be called a DR

[simsurace]

For example, if one tries to capture a perfectly blank wall, no amount of "dynamic range" or pixels or format, large or small, will suffice simply because a photographic image can not be made in the first place as there is nothing for the system to focus on. Contrasts have to be present (and no, this is not related to AF failure, it is basic photography theory). There thus will be a single value of whatever statistic one implements as a measure of "dynamic range" as at it is not determinable it can be set at any value one wishes as it cannot be changed later. It is a singularity.

The example you are giving is interesting, because it suggests that you are thinking of a per-image definition of DR. I don't think that this would be useful. If we are talking about properties of the scene, we should IMO use the term 'scene contrast' to describe the range of tones that is present in the scene. DR applies to the capture device, either on a per-photosite basis or on a per-area or per-image basis, depending on the question.

We would not measure DR by trying to record a scene with no scene contrast. We present a scene with excessive scene contrast and then determine to what extent this can be reproduced by the capture device. Depending on the question, we then normalize the measurement to a fixed output size, or we don't. The resulting number gives an upper bound to the range of tones that can be reproduced with an ideal output medium. On a limited output medium, the bottleneck changes from the capture device to the output medium. This does not invalidate the concept of DR of the capture device, in a way that is analogous to other concepts like resolution.

[Bjørn Rørslett]

Normalisation removes degree(s) of freedom. Instead assumptions are inserted that might not apply for the person trying to use the model. All of this makes it quite hard to come up with testable predictions of any practical value. I leave the scene there. So many pitfalls and loopholes to attend to. The theorists should start removing any of the many sources for circular reasoning. That'll save us a lot of time in futile discussions, for sure.

[simsurace]

Normalisation removes degree(s) of freedom. Instead assumptions are inserted that might not apply for the person trying to use the model. All of this makes it quite hard to come up with testable predictions of any practical value. I leave the scene there. So many pitfalls and loopholes to attend to. The theorists should start removing any of the many sources for circular reasoning. That'll save us a lot of time in futile discussions, for sure.

Simplification requires assumptions. There is no magic in this. But the various descriptions can happily coexist, I don't see a major danger involved. People that use the concept should first familiarise themselves with it and be aware of the assumptions, but this holds for any field or topic.

[bclaff]

Normalisation removes degree(s) of freedom. Instead assumptions are inserted that might not apply for the person trying to use the model. All of this makes it quite hard to come up with testable predictions of any practical value. I leave the scene there. So many pitfalls and loopholes to attend to. The theorists should start removing any of the many sources for circular reasoning. That'll save us a lot of time in futile discussions, for sure.

Normalization allows "apples to apples" comparisons and with further appropriate math can yield "oranges to oranges" as well.
In My Opinion (IMO), that far better than "apples to oranges" without normalization.

Not quite on topic but I'd like to point out that I do have per pixel measures such as Read Noise and Input-referred Read Noise at my site.
(As well as investigations of Fixed Pattern Noise (FPN) visualized as heatmaps.)

Naturally, whenever one consults technical data, it's important to choose the right data given the needs.
I think PDR is a very valuable measure; but certainly not appropriate for all discussions.

[David H. Hartman]

... This is notwithstanding the obvious logical flaws of requiring a *standard size output* (most photographers would never think this way anyway)...

This is for test purposes only. Why is it a logical flaw?

[Andrea B.]

Again, having the same output photo dimensions is only for the purpose of determining comparative gear characteristics. It has nothing to do with the photographic end purpose.

Two.Separate.Goals. -----> Compare Gear Characteristics versus Make Photographs.

Why Bjørn persists in conflating the two goals, I don't know.

[Les Olson]

[quote author=simsurace link=topic=5905.msg95526#msg95526 date=1495027979
- Les Olson argues (again, if I understood him, if not, please correct me and provide clarification) that all DR measures are useless
[/quote]

Photography is, as John Szarkowski put it, a process of visual editing: of keeping some things and leaving others out.  Everything outside the frame and everything before and after the image is made is left out.  The fact that in some cases, a bit of shadow or highlight detail is also left out is just part of the same process.  The idea that more sensor dynamic range means better photographs because less is left out is as inane as the idea that colour is better than B&W or video is better than still photography because less is left out.

Of course, there is such a thing as photography that is instrumental rather than aesthetic, where, it is easy to suppose, more dynamic range would always be better.  But for many such uses technical demands are low: telemedicine, eg, where phone images are perfectly acceptable for diagnosis of skin lesions.  There are also areas, such as anatomy, where photographs cannot compete with drawings despite their vastly superior resolution and dynamic range.

[Les Olson]

This is for test purposes only. Why is it a logical flaw?

Because the test is supposed to be a test of the gear for photography, and requiring a standardised output means it isn't.

[simsurace]

[quote author=simsurace link=topic=5905.msg95526#msg95526 date=1495027979
- Les Olson argues (again, if I understood him, if not, please correct me and provide clarification) that all DR measures are useless


Photography is, as John Szarkowski put it, a process of visual editing: of keeping some things and leaving others out.  Everything outside the frame and everything before and after the image is made is left out.  The fact that in some cases, a bit of shadow or highlight detail is also left out is just part of the same process.  The idea that more sensor dynamic range means better photographs because less is left out is as inane as the idea that colour is better than B&W or video is better than still photography because less is left out.

Of course, there is such a thing as photography that is instrumental rather than aesthetic, where, it is easy to suppose, more dynamic range would always be better.  But for many such uses technical demands are low: telemedicine, eg, where phone images are perfectly acceptable for diagnosis of skin lesions.  There are also areas, such as anatomy, where photographs cannot compete with drawings despite their vastly superior resolution and dynamic range.

I understand your argument, but I was never making any claims regarding artistic importance. More dynamic range does not equal better photographs. More dynamic range equals an objectively and technically better tool to record light signal, in terms of quantifiable, measurable criteria. It has nothing to do with art.

It's a similar question as 'how many keys does a piano have'. More keys provide a bigger frequency range to make music with. It does not equal 'better' music, nor do you have to use all the keys in every piece you play on that instrument.

Is there, in your opinion, any instance where more dynamic range in the capture device would be a disadvantage?

Many photographers are finding that digital cameras have vastly improved since their inception. You will not find many that want to go back to the first cameras with their rather small dynamic range. And they will continue to get better. This is just an opportunity for artistic expression, not a guarantee that artistic expression will become better in any way (this is a different topic altogether, and I wonder why it is so difficult to distinguish between the two).

Soon, we will get new output devices (screens) that are able to display more of the dynamic range that we are able to record. This will allow us to see photographs in a way that was not possible before. Some people will be head over heels about it, others will find it irrelevant for their artistic goals. Tastes differ.

[bclaff]

... This is notwithstanding the obvious logical flaws of requiring a *standard size output* (most photographers would never think this way anyway),...

FWIW, my Photographic Dynamic Range (PDR) does not specific a standard size output although the Signal to Noise Ratio (SNR) criteria is phrased using a specific example.
The measurement does assume a "typical" ratio of viewing distance to output size. It covers any size by applying the principle of similar triangles.
Most users of the chart would find the PDR value (on the y-axis) that meets their personal requirements by following the curve for a camera with which they are familiar up the the highest ISO setting for which they tend to get good results.
This process accounts for typical viewing distance and size for that photographer.
Comparing to other cameras at the same PDR level is sensible (to me and many photographers have reinforced this in their comments).

Although only one ratio of viewing distance to output size is measured, other ratios would not move thee relative positions of the sensors.

In many ways the central principle of PDR relates to the Circle Of Confusion (COC).
I haven't seen any intense debates about the use of COC in the computation of Depth Of Field (DOF).
To me taking PDR to task over the choice of using COC is roughly equivalent.

BTW, I'm only defending PDR here. I think DxOMark has made a serious error in their normalization strategy.