The GetDPI Photography Forum

Great to see you here. Join our insightful photographic forum today and start tapping into a huge wealth of photographic knowledge. Completing our simple registration process will allow you to gain access to exclusive content, add your own topics and posts, share your work and connect with other members through your own private inbox! And don’t forget to say hi!

Testing Real World Dynamic Range

Dogs857

New member
Before I get started this is not designed to start a war. Nor is it designed to be a "mine is better than yours" discussion. Also I am not a professional tester, I do this for my own benefit and would encourage everyone to test their own sensors for their intended use rather than listen to me. I just thought this could be a good demonstration of the difference between theoretical and real world applications.

All photos were processed in Capture One. An LCC was applied and then everything was cropped and exported.

On forums around the world DXO is quoted often as the undisputed fact with regards to dynamic range. While this might be technically correct, I find it is rarely replicated in actual photography. My D3x had 13.7 stops according to DXO, however I don't recall ever taking a photo that captured that much information. So I was wondering how much of the reported 12.9 stops of my new P45+ I can actually use.
Here is my completely unscientific testing method, it is open to interpretation, errors, and is personally subjective.

You know, like photography :)

Step 1; Tape a test subject to a wall in an area with relatively even light. In this case my balcony (it's raining again here).

BTW I focused this using my Cambo loupe and GG. I am new at this and quite happy how it turned out. Taken with the Rodie 70mm.
Step 2; take a reading of the middle grey swatch (bottom left) and start there. Mine came out at 1/4s at ISO50 and f/8.
Then take a series of images above and below this reading. Crop out all the stuff you don't need and you end up with this;
1/250s

1/125s

1/60s

1/30s

1/15s

1/8s

1/4s

1/2s

1sec

2sec

4sec


Now here is the subjective part. I am looking for where the 18% grey and the white squares become indistinguishable for highlights, and the same with the black square for shadows. This will let me know what the DR of my sensor is for real world photography.

What about highlight and shadow recovery I hear you say. Well here they are;
2sec highlight +100

4sec highlight +100

1/250 shadow +25

1/500 shadow +100


So what do I see in all this.

Highlights I stop at 1sec, I can still see separation there so I am confident that my highlights won't blow. I could extend to 2sec (an extra stop) and use highlight recovery but I am uncomfortable with this as it leaves me no wiggle room at all. For the shadow end I stop at 1/125sec. I can pull detail out at 1/250 but it does introduce noise and once again leaves no wiggle room.
My target is large prints for sale, so I am after the best image quality I can get. If I was only printing small, or sharing on the web then I could go to the two extreme ends as the detail I lose, and the noise I introduce are inconsequential.
There is always NR software as well, however there is always a trade off with detail quality and this would be dependant on the subject photographed.

So for my intended use, the P45+ has a safe, usable 8 stops of DR. This means if I take a spot meter reading of the brightest part of my image, I can expect detail in all areas 8 stops below that. If the DR of the scene is greater than this I will need to use filters.

From my 12.9 stops of theoretical DR I can safely use 8, and this is in keeping with most cameras I have ever owned.
 

ErikKaffehr

Well-known member
Hi,

I would think that you could push up the exposure slider to 200 or even 400 procent and could get good detail in the dark rectangles. I don't think you are utilising the full dynamic range of your camera.

DxO-mark presents two curves, DR which measures the distance between dark noise and saturation. Tells essentially how much you can push shadows on an image exposed fully to the right. 3-4 stops are normally possible.

The tonal range curve shows a more conservative version of the same, assuming a decent amount of signal over noise in the dark areas. It is given in bits, which correspond to EV. On the P45+ it is 9.55 bits that is 9.5 EV.

Best regards
Erik


Before I get started this is not designed to start a war. Nor is it designed to be a "mine is better than yours" discussion. Also I am not a professional tester, I do this for my own benefit and would encourage everyone to test their own sensors for their intended use rather than listen to me. I just thought this could be a good demonstration of the difference between theoretical and real world applications.

All photos were processed in Capture One. An LCC was applied and then everything was cropped and exported.

On forums around the world DXO is quoted often as the undisputed fact with regards to dynamic range. While this might be technically correct, I find it is rarely replicated in actual photography. My D3x had 13.7 stops according to DXO, however I don't recall ever taking a photo that captured that much information. So I was wondering how much of the reported 12.9 stops of my new P45+ I can actually use.
Here is my completely unscientific testing method, it is open to interpretation, errors, and is personally subjective.

You know, like photography :)

Step 1; Tape a test subject to a wall in an area with relatively even light. In this case my balcony (it's raining again here).

BTW I focused this using my Cambo loupe and GG. I am new at this and quite happy how it turned out. Taken with the Rodie 70mm.
Step 2; take a reading of the middle grey swatch (bottom left) and start there. Mine came out at 1/4s at ISO50 and f/8.
Then take a series of images above and below this reading. Crop out all the stuff you don't need and you end up with this;
1/250s

1/125s

1/60s

1/30s

1/15s

1/8s

1/4s

1/2s

1sec

2sec

4sec


Now here is the subjective part. I am looking for where the 18% grey and the white squares become indistinguishable for highlights, and the same with the black square for shadows. This will let me know what the DR of my sensor is for real world photography.

What about highlight and shadow recovery I hear you say. Well here they are;
2sec highlight +100

4sec highlight +100

1/250 shadow +25

1/500 shadow +100


So what do I see in all this.

Highlights I stop at 1sec, I can still see separation there so I am confident that my highlights won't blow. I could extend to 2sec (an extra stop) and use highlight recovery but I am uncomfortable with this as it leaves me no wiggle room at all. For the shadow end I stop at 1/125sec. I can pull detail out at 1/250 but it does introduce noise and once again leaves no wiggle room.
My target is large prints for sale, so I am after the best image quality I can get. If I was only printing small, or sharing on the web then I could go to the two extreme ends as the detail I lose, and the noise I introduce are inconsequential.
There is always NR software as well, however there is always a trade off with detail quality and this would be dependant on the subject photographed.

So for my intended use, the P45+ has a safe, usable 8 stops of DR. This means if I take a spot meter reading of the brightest part of my image, I can expect detail in all areas 8 stops below that. If the DR of the scene is greater than this I will need to use filters.

From my 12.9 stops of theoretical DR I can safely use 8, and this is in keeping with most cameras I have ever owned.
 

Dogs857

New member
Thanks for taking the time to have a look Erik.
9.5 seems about right. I will admit I am being conservative in how I am using this sensor but it guarantees me results.
How do you push the sliders to 400% when it only goes to 100?? (probably showing my lack of knowledge in capture one here). Noise is the issue that concerns me when pushing shadows.
Next step is to take a well exposed image exposed TTR, taking a number of spot meter readings from my scene to see where the exposure values are, and seeing how this holds up in real world shooting.
Just gotta wait for the rain to stop.
 
For shadow details you can push the "shadow" bar in the "high dynamic range" in the "Exposure" tab Capture One. If you reduce "contrast" at the same time you would get more shadow details than this, though you would get more noise (lower SNR) for underexposed shots. What SNR in shadow is acceptable is personal.

For highlight recoverability I would recommend shooting a colorchecker passport, if you are interested in real world use cases instead of digging pixel data through Raw Digger. Sometimes if one or two channels out of R, G, B ,G2 are blown out, Capture One can still guess the color, but only for simple objects such like your testing target. You would need a colorchecker passport to do sanity check for the highlight recoverability.
 
Sometimes Capture One can manage to guess the color for partially blown pixels on simple objects, giving false impression that the highlight recoverability is awesome. However for complex objects this is not always true.

 

Dogs857

New member
Cheers Void

Yeah the noise is the issue with the sliders. The 1/500 at +100 shows this, and for my use this is not acceptable, which is why I said it was subjective. If I wasn't going to sell it, or was putting something on the web then I can push this friendship a lot further.

I was quite impressed with the highlight recovery at 2secs, I was sure this was blown out. Once the rain moves on a bit I can try this out a bit more. I will look into the colour checker, have never considered one an option before but you never know.
 

jerome_m

Member
Before I get started this is not designed to start a war. Nor is it designed to be a "mine is better than yours" discussion. Also I am not a professional tester, I do this for my own benefit and would encourage everyone to test their own sensors for their intended use rather than listen to me. I just thought this could be a good demonstration of the difference between theoretical and real world applications.
This is actually a very good idea and I will use that test protocol myself.
 
A good way to evaluate whether dxomark gives estimation of real-world use cases is to download RAW files from dpreview's new studio scene comparison tool. Note that the exposure for the IQ180 there is overexposed so it may not be a straight-forward comparison without Raw Digger analysis.

Both the D3X and the P45+ are older models hence not included in the new tool. D3X can be found in the old studio comparison tool, but P45+ was not tested. I picked the 645D instead (similar Kodak CCD?).

If you download the RAW files and abuse them in ACR (as C1 does not support Pentax medium format digital), you would be able to evaluate SNR in deep shadow. As can be seen, the newer Sony CMOS (D600) indeed beats the older Sony CMOS (D3X) in shadow SNR, which is consistent with the dxomark numbers. The 645D still seems to be a bit noisy even down-sampled to the same size of the D3X, but as the color cast of the D3X seems to be more severe I would say that the shadow of the 645D wins here. But actually when you do analysis with Raw Digger you would be able to see that the 645D is overexposed when compared against the D3X. Also the D3X shot with 12-bit RAW which would be less than ideal for DR analysis.

 

Bryan Stephens

Workshop Member
The a7r and the D810 seem to be pretty similar except for their color profiles.

The canon is just ...... :facesmack:

I hope that the 5DSR sensor is better than that, or I may be switching out of Canon.
 

Chris Beards

New member
Hi Dogs, Nice straight forward test for simple souls like me who have spent hours zone testing FP-4/HC-110 development combinations. Most of my work is on 5x4 FP-4 as I still love the darkroom and a big palpable negative I can hold in my hand.
However I also have a P45+ and Techno which I have yet to get to grips with as I don't have a printer to access the output, do you make prints of the test patches for final assessment as with film testing.
 

Dogs857

New member
Hey Chris

Yes mate I do a print out of all the targets for evaluation. (though mine is all digital, I would be lost if you put me in a darkroom :) )I also take a spot meter reading between each exposure to make sure the base value hasn't changed. This is very basic testing but I find it quite useful.
 

Shashin

Well-known member
Sorry, Jeff, but your test makes no sense--changing exposure just does not move the tonal range up and down the scale, it expands and contracts it (the difference among the patches changes). Why not go out on a bright day and measure different points in a scene with a spot meter and then take a picture based on the metered exposure? That will tell you more than what you just did.

Print quality at any size has nothing to do with DR. High quality prints have to do with your skill as a printer (and whether you can expose correctly). DR has nothing to do with that.

BTW, DXO mark scores are "real world." They use exactly the same photons you use in Australia. Since you have no idea what the actual DR of a scene is (you cannot tell by simply looking at a picture (or even the scene)), how do you know if the controlled test by DXO Mark is right or wrong? I can tell you one thing, I can get more than eight stop out of my 645D and p25+ and a whole host of digital cameras I have used--there is no comparison to slide film that was probably eight stops.
 

ErikKaffehr

Well-known member
Re: Testing Real World Dynamic Range (some examples)

Hi,

I take the liberty to post an image with a large dynamic range, although this is an MFD forum and that shot was taken on Sony:

Many processing variants are possible, like this one:


The second one shows a couple of canoeists:


The same part in "RawDigger" is shown below. The red channel is at around -7EV. As a side note, that sensor probably holds 60000 electron charges at saturation and the range goes to 16000, so each number value corresponds to about four captured photons. So the number of 20 means around 80 photons. Shot noise would give an SNR (Signal Noise ratio of about nine). The peak is still a nice gaussian, indicating that it is not affected by readout noise.


The area around the sun is clearly clipped in some channels and cannot be recovered correctly but the disc of the sun is clearly visible.


The final image shows the whole picture in RawDigger. I would suggest that DR in this image is mainly limited by lens flare.




Best regards
Erik
 
Last edited:
Sorry, Jeff, but your test makes no sense--changing exposure just does not move the tonal range up and down the scale, it expands and contracts it (the difference among the patches changes). Why not go out on a bright day and measure different points in a scene with a spot meter and then take a picture based on the metered exposure? That will tell you more than what you just did.

Print quality at any size has nothing to do with DR. High quality prints have to do with your skill as a printer (and whether you can expose correctly). DR has nothing to do with that.

BTW, DXO mark scores are "real world." They use exactly the same photons you use in Australia. Since you have no idea what the actual DR of a scene is (you cannot tell by simply looking at a picture (or even the scene)), how do you know if the controlled test by DXO Mark is right or wrong? I can tell you one thing, I can get more than eight stop out of my 645D and p25+ and a whole host of digital cameras I have used--there is no comparison to slide film that was probably eight stops.
Really? :ROTFL:

You can shoot the above image (which ErikKaffehr posted #14 of this thread) with a 645D and claim that print quality at any size has nothing to do with DR? :ROTFL::ROTFL::ROTFL:

Seriously, you must have some decent and efficient techniques to deal with alignment issues I believe ;):rolleyes:
 

jlm

Workshop Member
test seemed reasonable to me. i used to do the same thing with tri-x a la fred picker, used a terry cloth towel for a target and made a range of exposures, looking for the white and black to just lose detail
 

Dogs857

New member
Sorry, Jeff, but your test makes no sense--changing exposure just does not move the tonal range up and down the scale, it expands and contracts it (the difference among the patches changes). Why not go out on a bright day and measure different points in a scene with a spot meter and then take a picture based on the metered exposure? That will tell you more than what you just did.

Print quality at any size has nothing to do with DR. High quality prints have to do with your skill as a printer (and whether you can expose correctly). DR has nothing to do with that.

BTW, DXO mark scores are "real world." They use exactly the same photons you use in Australia. Since you have no idea what the actual DR of a scene is (you cannot tell by simply looking at a picture (or even the scene)), how do you know if the controlled test by DXO Mark is right or wrong? I can tell you one thing, I can get more than eight stop out of my 645D and p25+ and a whole host of digital cameras I have used--there is no comparison to slide film that was probably eight stops.
Hey mate

Thanks for the input. I agree that the tonal range will change, I am only interested in when I can see the change in highlights and shadows. It's why I take a reading between each shot. So I know that if I take a reading off my highlights and it comes out at 1/60 I can expose at 1/15 (1/8 in a pinch) and not clip the highlights. I get 2-3 stops above middle, and 5 stops under. This is my safe range, I can get more but it takes more work and introduces noise that I find unacceptable. YMMV.
This is really only step one, the next is to take some samples and confirm what I have just done. The test is a bit less hit and miss than taking a series of test shots and then trying to remember what EV each part of the scene was. By doing this test first I can looks for my 8 stops and see how it pans out. If I can get more great, but 8 will be what I am expecting to see.

Agreed there is so much more that goes into making quality prints. This is designed to get me the best exposure for my scene, and to be able to expect what I can and can't expose for. It is one of many steps in the capture to print process.

This is by no means a technical test, I'm sure there are better and more accurate ways to do this, but it's simple and more importantly works for me.

Yes the DXO mark tests are the same worldwide, however it is a technical test that measures the photon sites. Can you really get 12+ stops of detail in a print from the 645D?? I'm not arguing that the DXO tests are wrong, but maybe not practical (I can see that statement sending this thread south) for me.

I will take some photos over the coming week and will post them up with EV ratings for various areas of the scene and see how this all goes. I may well just shoot myself in the face, or it may actually prove valid. All I know is it has worked pretty well before.
 

Shashin

Well-known member
Really? :ROTFL:

You can shoot the above image (which ErikKaffehr posted #14 of this thread) with a 645D and claim that print quality at any size has nothing to do with DR? :ROTFL::ROTFL::ROTFL:

Seriously, you must have some decent and efficient techniques to deal with alignment issues I believe ;):rolleyes:
Voidshatter, you keep confusing technical specifications with photography.
 

Shashin

Well-known member
test seemed reasonable to me. i used to do the same thing with tri-x a la fred picker, used a terry cloth towel for a target and made a range of exposures, looking for the white and black to just lose detail
How did you determine the values of the white and black?

I have tested film as well. The way to do this is photograph (or, in my case, print a known densitometry strip) a target of known values and range. Then you measure with a densitometer. You need to then plot those values. The second step would be to figure out the rest of the reproduction process.

Now, perhaps the test gave you information you can somehow visualize in your process. But how do you then use it? You cannot tell the scene contrast by simply looking at it. But to then say that test showed a DR of X stops is one step too far.

All I am really saying is that I am not sure what Jeff really has. If he was taking only one patch and changing exposure, that would make sense, but he is taking two, which does not. If he is looking for the exposure latitude, which is different from DR, then I can see that two patches would show him how many stops in this particular situation would give him, but a different scene contrast would change that result.

I just think it is simpler and more useful to measure scene contrast and then photograph it and compare the two. But to be honest, the best way is through experience where a photographer learns to work intuitively. It would be a really pain in the neck to measure scene contrast and then try to figure out exposure each time (This is why Adams made a subjective scale (which it not accurate) to get photographers to visualize a scene for the Zone System. And the fact that scene contrast exceeds DR does not prevent a great image to be taken. Photographers have always had that problem and solved it by controlling exposure and reproduction.
 
Top