Discussion of 'Equivalence'

[simsurace]

Yes?? and no. The claimed increase is a direct function of N, the number of pixels. Thus if the density of pixels is increased on SAME sensor, the value will increase.

If the density increases, then the full well capacity usually goes down. Otherwise you would require more exposure to fill the well. If you keep full well capacity the same while decreasing pixel size, then you got yourself a lower base ISO. The higher DR would again make sense.

[Bjørn Rørslett]

I exemplified by using subsampling from a sensor. No change in well capacity as the data are already processed in-camera.

The 'theory' is hardly universal if it only applies to very specific situations, or not? That apart from the questionable practice of heavily massaging the data.

Again, the entire debate is going around in circles. At this stage I'm tempted to ask say the Fuji fans why they use that system because 'theory' says it is inferior ?? Perhaps practice indicates otherwise? I really hope and believe so.

[Jakov Minić]

Bjørn posted a screen shot that baffles me and I believe it's fiction.
How can it be that if you use the D5 in DX mode you lose dynamic range?
I have been shooting DX lenses (10.5/2.8 fish-eye and 40/2.8 micro) in FX mode on D4, Df, and D750.
Was I increasing dynamic range just because I shot in FX mode??? Would I have lost dynamic range had I shot in DX mode???
I don't get it!

[simsurace]

I exemplified by using subsampling from a sensor. No change in well capacity as the data are already processed in-camera.

The 'theory' is hardly universal if it only applies to very specific situations, or not? That apart from the questionable practice of heavily massaging the data.

If you are subsampling, you are throwing away the photons that were recorded in the discarded photosites. Statistically speaking, it has the same effect as using photosites that are less efficient, or increasing the ISO and exposing less.

The theory is certainly not universal: it does not account, among other things, for destructive operations on the data like your subsampling strategy. As an extreme example, painting all the pixels black would set the DR to zero irrespective of everything else  . One can hardly say that this would be a failure of 'equivalence'.

[Bjørn Rørslett]

Whatever floats your boat.

I think I have gotten the answers I need at this stage. One cannot alter the already sampled data by subsampling later. A 'theory' that puts 'response' before 'action' implies use of negative time. If one is content with such nefarious miracles, be my guest.

[simsurace]

Bjørn posted a screen shot that baffles me and I believe it's fiction.
How can it be that if you use the D5 in DX mode you lose dynamic range?
I have been shooting DX lenses (10.5/2.8 fish-eye and 40/2.8 micro) in FX mode on D4, Df, and D750.
Was I increasing dynamic range just because I shot in FX mode??? Would I have lost dynamic range had I shot in DX mode???
I don't get it!

Hi Jakov, thanks for voicing your concerns!

The PDR is a measure of the noisiness of images that result from a certain sensor. Cameras with the same PDR should produce images that are similarly noisy.
When you use only a DX-sized portion of the FX sensor, it is similar as if you are using a native DX sensor of the same technology.
Of course, if the images are otherwise not similar at all, you will not be able to observe this.
But if you use your FX cam in crop mode and set everything in order to get a similar image, you will probably see this in effect.

As an example:

Shoot the same scene from the same position (entrance pupil location) using:
1) a 150mm lens on FX, exposed at 1/100s f/8 at ISO3200
2) a 100mm lens in DX crop, exposed at 1/100s f/5.6 at ISO1600

The two resulting images will look very similar and have similar noise when viewed at the same output size. The fact that the DX crop has a lower PDR results in getting the same noise as the FX image at half the ISO setting.

[simsurace]

Whatever floats your boat.

I think I have gotten the answers I need at this stage. One cannot alter the already sampled data by subsampling later. A 'theory' that puts 'response' before 'action' implies use of negative time. If one is content with such nefarious miracles, be my guest.

I don't quite understand the purpose of an example where you deliberately throw away data after capturing it. What is this example supposed to prove? You can always make the image worse in some way, no general theory will account for that.

[Bjørn Rørslett]

Jakov: it's about the lens you put on your camera and how far you are away from your subject, ie. magnification and nothing else. Just go out and continue shooting without too much worries about these alleged "differences". They really exist only in the eye of he beholder.

[Bjørn Rørslett]

I don't quite understand the purpose of an example where you deliberately throw away data after capturing it. What is this example supposed to prove? You can always make the image worse in some way, no general theory will account for that.

To illustrate the fallacies in the suggested model. DR here is only a function of N, number of included pixels. Which is of course pretty nonsensical.

Had the 'theory' had any applicability, going in either direction should have been taken care of.

So far, I haven't seen the slightest shred of evidence that support the claims of DR differences using the published graphs. I read through the more technical articles linked to and did not find anything relevant there.  Further quotations of authorities are superfluous as these cannot prove or disprove anything; hard data can. I have challenged people to come forth with such data and just observe a wall of silence - why?

A final word about "throwing away data". In printing to a fixed size, at least some if not all cameras require one to "throw away" pixels simply because the number recorded is higher than necessary. As today's better cameras can deliver huge prints much bigger than needed for print to magazines or newspapers, such "waste" of image data constitutes the norm. Not to speak of the low requirements for web images.  On the other hand, if a larger number of pixels would be required for the given print, interpolation have had to take place, again introducing possible artefacts.

[simsurace]

To illustrate the fallacies in the suggested model. DR here is only a function of N, number of included pixels. Which is of course pretty nonsensical.

Had the 'theory' had any applicability, going in either direction should have been taken care of.

Why is it nonsensical?
John showed you a back-of-the-envelope estimation of DR of a sensor composed of pixels of a given per-pixel DR.
Of course it is not accurate if you deliberately process the data to reduce DR.

[Bjørn Rørslett]

I simply cannot accept the statement unless it is evidenced by hard data, that's why. Obviously a tough challenge to comply with?

Please, if this debate is going to be at all useful, new and better arguments need to be made, not by putting forth predictions from a model. A model cannot prove anything per se.

I fear the "science" here is unlikely to have survived peer reviewing in any decent journal.

[Andrea B.]

oh heaven have mercy!

One of you is making arguments based on the (Generally Accepted) Definition of "Equivalence". One of you is not. Neither of you is wrong.

For a proper debate, you both must agree on the definition of terms beforehand.

As per Joseph James in the link above we have that "Equivalent" photos are photos of a given scene all of which have the following characteristics.

•     Same Perspective
•     Same Framing
•     Same DOF, Diffraction, Total Amount of Light on the Sensor
•     Same Exposure Time
•     Same Brightness
•     Same Display Dimensions

This is a very specialized sub-definition for the common word 'equivalence'.

As per this specialized Definition of "Equivalence", we can observe that Bjørn's example -- Foto1 shot as FX and Foto2 cropped from Foto1 to make a derived DX -- are NOT "Equivalent" because they do not have the Same Framing.

Foto1 and Foto2 quite do have the same perspective, DOF, Diffraction, Exposure Time, Brightness. (As they are theoretical, we don't know if Foto1 and Foto2 have the same display dimensions, but who cares anyway.) 


Whether or not one likes this Definition of "Equivalence" is beside the point because that is how "Equivalence" is currently defined and used in the various photographic arguments across the web. However I seriously doubt whether most users of the term actually understand the implications and proper usage of it when making their arguments. And language barriers get in the way also.

[simsurace]

So far, I haven't seen the slightest shred of evidence that support the claims of DR differences using the published graphs. I read through the more technical articles linked to and did not find anything relevant there.  Further quotations of authorities are superfluous as these cannot prove or disprove anything; hard data can. I have challenged people to come forth with such data and just observe a wall of silence - why?

Well, I have only seen you ask for the test of one specific hypothesis, which I did not put forward. I hope it is clear why I'm unlikely to engage in such an exercise.
Too busy on testing hypothesis that I did in fact put forward. 

I'm not sure what portion of my explanations of the definition of PDR was unclear.
The graphs of PDR only make sense if you accept what PDR is.
Would you instead endorse a DR definition that uses the same SNR cutoff for all formats? Why?
If yes, would this make it easier to interpret the resulting graphs in terms of predicting photographic output of a fixed size?

[Bjørn Rørslett]

The properties of 'Equivalence' (Andrea)  quotes can only be attained by using the same camera and lens all the time. It actually defines repeated sampling of a scene, nothing else.

It was bad before. Now it is getting worse. No offence .

I officially give up. What have been learnt? Nothing that wasn't already known in the 19th Century, way ahead of the digital era.

[simsurace]

I simply cannot accept the statement unless it is evidenced by hard data, that's why. Obviously a tough challenge to comply with?

Please, if this debate is going to be at all useful, new and better arguments need to be made, not by putting forth predictions from a model. A model cannot prove anything per se.

I fear the "science" here is unlikely to have survived peer reviewing in any decent journal.

If you are thinking about John's argument, I think an introductory text about Poisson processes would be a start.
Unless you are asking me to experimentally re-prove the Poisson statistics of photon arrivals, I don't see what kind of experiment is needed to support this kind of standard calculation.

Peer reviewing is basically what we are doing here. Despite the obvious absence of the authors.

EDIT: Sorry, but I got confused here. John's argument does not rely on Poisson statistics.