DR testing results

[A74Me]

I have downladed and pushed the shadows of test images from the canon d30 and the results were nothing short of amazing.











pushed 4 stops

D30



GFx 100

 

[Ruud uit de Graaf]

Hi A74,

did you downsize (resample) or just crop part of the attached pictures for us?

What do you mean by amazing? What amazes someone is based on that individuals expectations. What where your expectations?

thanks,

Ruud

 

[A74Me]

Hi A74,

did you downsize (resample) or just crop part of the attached pictures for us?

What do you mean by amazing? What amazes someone is based on that individuals expectations. What where your expectations?

thanks,

Ruud

just cropped, that the testing method is , well.... not quite right, wouldnt you say.

 

[Ruud uit de Graaf]

Hi A74,

did you downsize (resample) or just crop part of the attached pictures for us?

What do you mean by amazing? What amazes someone is based on that individuals expectations. What where your expectations?

thanks,

Ruud

just cropped, that the testing method is , well.... not quite right, wouldnt you say.

I would say, it doesn't make sense to compare 1:1 views of camera's with complete different sensor sizes ...

and 2nd : the testcharts / pictures /methods have changed significant between the 2 camera's tested, so that makes it unreliable as well.

Ruud

 

[A74Me]

Hi A74,

did you downsize (resample) or just crop part of the attached pictures for us?

What do you mean by amazing? What amazes someone is based on that individuals expectations. What where your expectations?

thanks,

Ruud

just cropped, that the testing method is , well.... not quite right, wouldnt you say.

I would say, it doesn't make sense to compare 1:1 views of camera's with complete different sensor sizes ...

and 2nd : the testcharts / pictures /methods have changed significant between the 2 camera's tested, so that makes it unreliable as well.

Ruud

formular

2000 = 8 stops

2025 = 9.5 stops + marketing = 15 stops

 

[Ruud uit de Graaf]

A74Me wrote:
formular

2000 = 8 stops

2025 = 9.5 stops + marketing = 15 stops

Where does this come from? From your experiment? :-o



I don't think this thread should belong to the Science en technology section

 

[John Sheehy]

I have downladed and pushed the shadows of test images from the canon d30 and the results were nothing short of amazing.

What is amazing about heavy NR on solid rectangles where the NR has no fine detail to destroy?

Now, try the two raw files with a converter that doesn't use any NR. Most popular converters use NR even when you tell them NR should be zero, because the public would not accept the actual noise of cameras in shadows and at high ISOs. "No NR" often means "Only standard NR, with no boost".

 

[Richard Butler]

formular

2000 = 8 stops

2025 = 9.5 stops + marketing = 15 stops

This is not what you've demonstrated, and there are a number of reasons you can't draw strong conclusions from what you've shown. Let's put aside that these shots are taken under different lighting with very different exposures.

It's also unclear whether you're working from Raw or JPEG (JPEGs typically have a fixed S-curve applied and are then reduced to an 8-bit space, discarding a lot of the DR in the process, so can't be used for assessing DR in the sense that Photons to Photos measures it).

The amount of each image clipping to white doesn't tell you anything about DR

My concern is that you're trying to interpret DR based on how much of the images have clipped, which doesn't tell you anything at all about DR.

Dynamic range is the range between the point of clipping and the darkest usable tone. (The definition used by engineers is typically a signal/noise ratio of 1. Bill Claff's data uses a different cut-off).

- It does not specify where middle grey is

- Given you work back from a hard clipping point, differences in DR are found in the shadows.

Two cameras with radically different DR levels may still devote the same amount of their captured DR to the point between clipping and middle grey. So the degree to which these two images clip to white doesn't tell you anything useful about DR, other than that they have similar clipping-to-mid grey ranges (which tells you more about the JPEG tone curves of the cameras than about their sensors).

Differences in the shadows

There are no deep shadows in the initial images, so they're not going to be able to tell you much about DR differences, no matter how you process them. But the dark sub-midtones (patches 17, 18 and 19 on the Kodak chart) are significantly cleaner on the GFX than the D30, which strongly suggests the GFX will have much cleaner tones as you push further and further into the shadows, consistent with it having more DR.

These shots don't let you assess that in any real detail, but nothing in them supports your implication that DR numbers are simply marketing claims.

Richard - DPReview.com

 

[A74Me]

formular

2000 = 8 stops

2025 = 9.5 stops + marketing = 15 stops

This is not what you've demonstrated, and there are a number of reasons you can't draw strong conclusions from what you've shown. Let's put aside that these shots are taken under different lighting with very different exposures.

It's also unclear whether you're working from Raw or JPEG (JPEGs typically have a fixed S-curve applied and are then reduced to an 8-bit space, discarding a lot of the DR in the process, so can't be used for assessing DR in the sense that Photons to Photos measures it).

The amount of each image clipping to white doesn't tell you anything about DR

My concern is that you're trying to interpret DR based on how much of the images have clipped, which doesn't tell you anything at all about DR.

Dynamic range is the range between the point of clipping and the darkest usable tone. (The definition used by engineers is typically a signal/noise ratio of 1. Bill Claff's data uses a different cut-off).

- It does not specify where middle grey is

- Given you work back from a hard clipping point, differences in DR are found in the shadows.

Two cameras with radically different DR levels may still devote the same amount of their captured DR to the point between clipping and middle grey. So the degree to which these two images clip to white doesn't tell you anything useful about DR, other than that they have similar clipping-to-mid grey ranges (which tells you more about the JPEG tone curves of the cameras than about their sensors).

Differences in the shadows

There are no deep shadows in the initial images, so they're not going to be able to tell you much about DR differences, no matter how you process them. But the dark sub-midtones (patches 17, 18 and 19 on the Kodak chart) are significantly cleaner on the GFX than the D30, which strongly suggests the GFX will have much cleaner tones as you push further and further into the shadows, consistent with it having more DR.

These shots don't let you assess that in any real detail, but nothing in them supports your implication that DR numbers are simply marketing claims.

Richard - DPReview.com

DPR quote: The first thing that strikes you about the D30's images is their virtual lack of any noise (obviously some becomes apparent at higher ISO's but still impressive)

form what i remember bcak in the early 2000 DR was measured bu not manipulating the image file and its was always around 8 stops. but today the files are push to the max and then NR is applied then dR is measured.

shadows lifted effortlessly



original file posted by Dpr

 

[A74Me]

formular

2000 = 8 stops

2025 = 9.5 stops + marketing = 15 stops

This is not what you've demonstrated, and there are a number of reasons you can't draw strong conclusions from what you've shown. Let's put aside that these shots are taken under different lighting with very different exposures.

It's also unclear whether you're working from Raw or JPEG (JPEGs typically have a fixed S-curve applied and are then reduced to an 8-bit space, discarding a lot of the DR in the process, so can't be used for assessing DR in the sense that Photons to Photos measures it).

8 stops from jpeg is not disgarding more than the tolerance of electrical components, from what i understand, even though i found this artical interesting regarding op amps,

Operational amplifiers

"Op-amps are a whole other story. Inside the plastic package is a full circuit consisting of transistors, resistors and capacitors (hence the name “integrated circuit” or IC). They’re designed in such a way that the performance of the IC as a whole is very consistent from unit to unit despite the variance in the individual parts inside.

There are still several factors such as noise level or input offset current that can vary. However, the major distinction with op-amps is that most datasheet specs have a “typical” value, which is what you can expect most of them to test at or near. This is very different than just giving a range without any guidance as to where in the range you can expect it to be.

Plus, with op-amps, the types of specs that do vary piece to piece are typically not important for low-level audio work. The actual performance of a modern op-amp is predictable and precise without any biasing or calibration. If you put a 100k resistor in the feedback loop and a 10k from inverting input to Vbias, it’s going to give you a gain of 10. The tolerance of those two external resistors will have a much bigger impact on the circuit than anything going on inside the chip."

bills diagram shows 2x the DR difference

The amount of each image clipping to white doesn't tell you anything about DR

My concern is that you're trying to interpret DR based on how much of the images have clipped, which doesn't tell you anything at all about DR.

Dynamic range is the range between the point of clipping and the darkest usable tone. (The definition used by engineers is typically a signal/noise ratio of 1. Bill Claff's data uses a different cut-off).

- It does not specify where middle grey is

- Given you work back from a hard clipping point, differences in DR are found in the shadows.

Two cameras with radically different DR levels may still devote the same amount of their captured DR to the point between clipping and middle grey. So the degree to which these two images clip to white doesn't tell you anything useful about DR, other than that they have similar clipping-to-mid grey ranges (which tells you more about the JPEG tone curves of the cameras than about their sensors).

Differences in the shadows

There are no deep shadows in the initial images, so they're not going to be able to tell you much about DR differences, no matter how you process them. But the dark sub-midtones (patches 17, 18 and 19 on the Kodak chart) are significantly cleaner on the GFX than the D30, which strongly suggests the GFX will have much cleaner tones as you push further and further into the shadows, consistent with it having more DR.

These shots don't let you assess that in any real detail, but nothing in them supports your implication that DR numbers are simply marketing claims.

Richard - DPReview.com

 

[JimKasson]

I have downladed and pushed the shadows of test images from the canon d30 and the results were nothing short of amazing.







pushed 4 stops

D30



GFx 100

The phrase “not even wrong” is a critique in scientific and philosophical discussions. It means that a claim or idea is so ill-posed, incoherent, or misframed that it can’t even be evaluated; it fails to meet the minimum standard of being falsifiable. It was popularized by physicist Wolfgang Pauli.

A statement is “not even wrong” if:

• It’s vague, non-testable, or internally inconsistent
• It doesn’t make concrete predictions that could be checked
• It misuses terminology in ways that render it meaningless in context

I think Richard has explained the main issues.

--
https://blog.kasson.com

 

[A74Me]

I have downladed and pushed the shadows of test images from the canon d30 and the results were nothing short of amazing.







pushed 4 stops

D30



GFx 100

The phrase “not even wrong” is a critique in scientific and philosophical discussions. It means that a claim or idea is so ill-posed, incoherent, or misframed that it can’t even be evaluated; it fails to meet the minimum standard of being falsifiable. It was popularized by physicist Wolfgang Pauli.

A statement is “not even wrong” if:

• It’s vague, non-testable, or internally inconsistent
• It doesn’t make concrete predictions that could be checked
• It misuses terminology in ways that render it meaningless in context

I think Richard has explained the main issues.

explain these 2 black level comparrisions when pushed, the d30 clearly out performs the values in the shadows. the smaller pixels have crushed the information.

 

[Richard Butler]

I have downladed and pushed the shadows of test images from the canon d30 and the results were nothing short of amazing.







pushed 4 stops

D30



GFx 100

The phrase “not even wrong” is a critique in scientific and philosophical discussions. It means that a claim or idea is so ill-posed, incoherent, or misframed that it can’t even be evaluated; it fails to meet the minimum standard of being falsifiable. It was popularized by physicist Wolfgang Pauli.

A statement is “not even wrong” if:

• It’s vague, non-testable, or internally inconsistent
• It doesn’t make concrete predictions that could be checked
• It misuses terminology in ways that render it meaningless in context

I think Richard has explained the main issues.

explain these 2 black level comparrisions when pushed, the d30 clearly out performs the values in the shadows. the smaller pixels have crushed the information.

I've just found that motorbike picture in the review and I don't think we published the Raw file. Are you trying to assess DR from JPEGs?

Because, as I say, most standard JPEGs use tone curves to convey about 8.5 or so EV of dynamic range, discarding everything beyond that.

Trying to assess DR by pushing JPEGs is like smashing sets of crockery and trying to guess how maleable the clay they were made from was, before they were fired.

Richard - DPReview.com

 

[A74Me]

I have downladed and pushed the shadows of test images from the canon d30 and the results were nothing short of amazing.







pushed 4 stops

D30



GFx 100

The phrase “not even wrong” is a critique in scientific and philosophical discussions. It means that a claim or idea is so ill-posed, incoherent, or misframed that it can’t even be evaluated; it fails to meet the minimum standard of being falsifiable. It was popularized by physicist Wolfgang Pauli.

A statement is “not even wrong” if:

• It’s vague, non-testable, or internally inconsistent
• It doesn’t make concrete predictions that could be checked
• It misuses terminology in ways that render it meaningless in context

I think Richard has explained the main issues.

explain these 2 black level comparrisions when pushed, the d30 clearly out performs the values in the shadows. the smaller pixels have crushed the information.

I've just found that motorbike picture in the review and I don't think we published the Raw file. Are you trying to assess DR from JPEGs?

Because, as I say, most standard JPEGs use tone curves to convey about 8.5 or so EV of dynamic range, discarding everything beyond that.

Trying to assess DR by pushing JPEGs is like smashing sets of crockery and trying to guess how maleable the clay they were made from was, before they were fired.

Richard - DPReview.com

the histograms are near identical from a pushed gfx100 raw file vers jpeg, im wanting to see the extra 6 stops of DR demonstrated. to be honest id take the d30 jpeg over the raw gfx 100 file till someone can clearly show 6 stops difference when it starts behind from the start

 

[Ian Stuart Forsyth]

I have downladed and pushed the shadows of test images from the canon d30 and the results were nothing short of amazing.







pushed 4 stops

D30



GFx 100

The phrase “not even wrong” is a critique in scientific and philosophical discussions. It means that a claim or idea is so ill-posed, incoherent, or misframed that it can’t even be evaluated; it fails to meet the minimum standard of being falsifiable. It was popularized by physicist Wolfgang Pauli.

A statement is “not even wrong” if:

• It’s vague, non-testable, or internally inconsistent
• It doesn’t make concrete predictions that could be checked
• It misuses terminology in ways that render it meaningless in context

I think Richard has explained the main issues.

explain these 2 black level comparrisions when pushed, the d30 clearly out performs the values in the shadows. the smaller pixels have crushed the information.

I've just found that motorbike picture in the review and I don't think we published the Raw file. Are you trying to assess DR from JPEGs?

Because, as I say, most standard JPEGs use tone curves to convey about 8.5 or so EV of dynamic range, discarding everything beyond that.

Trying to assess DR by pushing JPEGs is like smashing sets of crockery and trying to guess how maleable the clay they were made from was, before they were fired.

Richard - DPReview.com

the histograms are near identical from a pushed gfx100 raw file vers jpeg, im wanting to see the extra 6 stops of DR demonstrated. to be honest id take the d30 jpeg over the raw gfx 100 file till someone can clearly show 6 stops difference when it starts behind from the start

First people have been trying to teach you the difference between the histogram you are showing as it is not a raw histogram and is of the processed raw data that has been placed with in a color space.

I know of at least 7 threads started over various platform that you have not learned the difference, until you fully understand what those extra stops of DR mean and how that relates to what is captured in the raw file I feel it will be the very same runaround many of us have dealt with the past 10 years and trying to tell you what that extra DR is and how it relates to how we process that data into the color space you are trying to display that data in.

In one final attempt to show you where this extra DR comes into play



In the red box the image data falls well below the DR that can be displayed in the color space in the tonal ranges that are held within the standard tonal range.

When we are discussing the DR of the sensor and how much is recorded, we are talking about the range of the lightest to the darkest data being collected with floor to the lowest signal determined by a noise level.



We are then compressing that range of DR into the color space we are going to display that data in. That red patch was well below the 0,0,0 color space with its placement in the raw data. We then compress that to how we want to display that tonal range of the raw data

--
The Camera is only a tool, photography is deciding how to use it.
The hardest part about capturing wildlife is not the photographing portion; it’s getting them to sign a model release

 

[A74Me]

I have downladed and pushed the shadows of test images from the canon d30 and the results were nothing short of amazing.







pushed 4 stops

D30



GFx 100

The phrase “not even wrong” is a critique in scientific and philosophical discussions. It means that a claim or idea is so ill-posed, incoherent, or misframed that it can’t even be evaluated; it fails to meet the minimum standard of being falsifiable. It was popularized by physicist Wolfgang Pauli.

A statement is “not even wrong” if:

• It’s vague, non-testable, or internally inconsistent
• It doesn’t make concrete predictions that could be checked
• It misuses terminology in ways that render it meaningless in context

I think Richard has explained the main issues.

explain these 2 black level comparrisions when pushed, the d30 clearly out performs the values in the shadows. the smaller pixels have crushed the information.

I've just found that motorbike picture in the review and I don't think we published the Raw file. Are you trying to assess DR from JPEGs?

Because, as I say, most standard JPEGs use tone curves to convey about 8.5 or so EV of dynamic range, discarding everything beyond that.

Trying to assess DR by pushing JPEGs is like smashing sets of crockery and trying to guess how maleable the clay they were made from was, before they were fired.

Richard - DPReview.com

the histograms are near identical from a pushed gfx100 raw file vers jpeg, im wanting to see the extra 6 stops of DR demonstrated. to be honest id take the d30 jpeg over the raw gfx 100 file till someone can clearly show 6 stops difference when it starts behind from the start

First people have been trying to teach you the difference between the histogram you are showing as it is not a raw histogram and is of the processed raw data that has been placed with in a color space.

I know of at least 7 threads started over various platform that you have not learned the difference, until you fully understand what those extra stops of DR mean and how that relates to what is captured in the raw file I feel it will be the very same runaround many of us have dealt with the past 10 years and trying to tell you what that extra DR is and how it relates to how we process that data into the color space you are trying to display that data in.

In one final attempt to show you where this extra DR comes into play



In the red box the image data falls well below the DR that can be displayed in the color space in the tonal ranges that are held within the standard tonal range.

When we are discussing the DR of the sensor and how much is recorded, we are talking about the range of the lightest to the darkest data being collected with floor to the lowest signal determined by a noise level.



We are then compressing that range of DR into the color space we are going to display that data in. That red patch was well below the 0,0,0 color space with its placement in the raw data. We then compress that to how we want to display that tonal range of the raw data

your red box is showing 19 19 19 its not below at all.

the d30 image lowest black is 12 12 12

the myth that the incamera histogram doesnt equal the raw , i disproved that 2 years ago. i just shot an image with the histogram just clipping the blacks and the reading was 0 1 1

there will never be true black in a high contrast scene ,lens veiling glare takes care of that.

 

[JimKasson]

the myth that the incamera histogram doesnt equal the raw , i disproved that 2 years ago.

If your assertion is that the in-camera histogram is the same as the raw histogram, that is easily disproven.

1. Put the camera on a tripod, set the white balance to 3200K.
2. Observe the histogram.
3. Now set the WB to 8000 K.
4. Observe that the histogram is different.
5. Examine raw files made with the two WB settings.
6. The raw data is the same.
7. QED.

 

[Richard Butler]

I've just found that motorbike picture in the review and I don't think we published the Raw file. Are you trying to assess DR from JPEGs?

Because, as I say, most standard JPEGs use tone curves to convey about 8.5 or so EV of dynamic range, discarding everything beyond that.

Trying to assess DR by pushing JPEGs is like smashing sets of crockery and trying to guess how maleable the clay they were made from was, before they were fired.

Richard - DPReview.com

the histograms are near identical from a pushed gfx100 raw file vers jpeg, im wanting to see the extra 6 stops of DR demonstrated. to be honest id take the d30 jpeg over the raw gfx 100 file till someone can clearly show 6 stops difference when it starts behind from the start

Is that image one that contains 6 additional stops of DR? How are we to tell?

What is your Raw converter of choice doing to the shadows? How would we know?

Presenting out-of-context screengrabs isn't a convincing way to disprove all existing theory and testing, especially if your reference point is a JPEG, which can't act as a meaningful point of comparison. Even more so when the first images you posted strongly suggested the GFX was capable of showing more dynamic range than the D30 (less noisy darkest tones).

If you download the Raw from this scene:

https://www.dpreview.com/reviews/fujifilm-gfx-50r-review/5#sbs

The brightest tones are at or very close to clipping and the darkest are around 13-14 stops below that. (ie: Raw values around 2 in the shadows and approaching 16813 in the highlights).

You can process that to have nothing rendered as quite black, and get a histogram with everything neatly showing in the middle of the graph, but that doesn't tell us anything. It certainly doesn't prove or disprove how much DR was captured.

Richard - DPReview.com

 

[knickerhawk]

the myth that the incamera histogram doesnt equal the raw , i disproved that 2 years ago.

If your assertion is that the in-camera histogram is the same as the raw histogram, that is easily disproven.

1. Put the camera on a tripod, set the white balance to 3200K.
2. Observe the histogram.
3. Now set the WB to 8000 K.
4. Observe that the histogram is different.
5. Examine raw files made with the two WB settings.
6. The raw data is the same.
7. QED.

Have you forgotten the discussion on DPRevived with A74Me a couple of years ago? Your demonstration there of your Hassy's WB-related histogram behavior and my Oly EM1iii red flower shot histogram demonstration held no sway over A74Me. His Sony A6300 live histogram proved it's all just a "myth"...

I fear we are beyond not even wrong. We are now deep into not even not even wrong! territory... ;-)

 

[A74Me]

the myth that the incamera histogram doesnt equal the raw , i disproved that 2 years ago.

If your assertion is that the in-camera histogram is the same as the raw histogram, that is easily disproven.

1. Put the camera on a tripod, set the white balance to 3200K.
2. Observe the histogram.
3. Now set the WB to 8000 K.
4. Observe that the histogram is different.
5. Examine raw files made with the two WB settings.
6. The raw data is the same.
7. QED.

Can you post some histogram images from your camera showing your results. both the RGB and brightness, then post the same histogram image from LiB raw with the highlight and show blinkies turned on ,also one from PS ACR allowing for the margin of error figure thats baked into ACR and if those are within .3 of a stop then the 2 test results are still 5.7 stops out .