EP1 Dynamic range - dxo says 10eV

[Eric_1]

I was comparing the dynamic range of my ep1 with my a850 on dxo - no surprises here the a850 wins by 1.5 eV at lowest iso - when I noticed that DxO claims the ep1 has 10 stops of dynamic range!

That doesn't match with my limited testing, which suggests the ep1 has only 6 stops of dynamic range. Am I testing incorrectly? Here's my set up:

I aim the ep1 at a white piece of paper at iso100. The live histogram shows a single narrow peak indicating most pixels are at the same brightness value. Next, I increase exposure value to its max (+3), and watch that little peak shift right until its mostly in the red - this occurs at +3. Next, I decrease exposure value to its min (-3) and watch that peak shift to the left until it's almost entirely in the blue (this occurs at -3 eV)

I interpret this to mean that there are 6 stops between the brightest and darkest pixels the ep1 can simultaneously capture. Am I wrong? How on earth could the ep1 possibly get 10 full stops of dynamic range? That'd be + -5 eV of exposure compensation! DxO must be measuring something different...
--
http://picasaweb.google.com/EricSL

 

[boyzo]

the EP-1 is a fine camera capable of excellent images just enjoy its capabilities ;-)
--
John

http://boyzo.smugmug.com
http://picasaweb.google.com/john.boyzo

 

[Kuivaamo]

DxO must be measuring something different...

No kidding. You can't use the JPEG output of a camera to determine the DR captured by the sensor, since the tone curve applied will obviously have a huge impact on the result. DxO use the RAW data.

With a proper RAW workflow you should easily get much more than 6 stops of DR from your E-P1.

 

[Eric_1]

I'm not complaining - I love the camera - I'm just curious. Not every question on these forums is a complaint

the EP-1 is a fine camera capable of excellent images just enjoy its capabilities ;-)
--
John

http://boyzo.smugmug.com
http://picasaweb.google.com/john.boyzo

--
http://picasaweb.google.com/EricSL

 

[Eric_1]

Are you saying the live histogram counts as jpeg? How would you actually test the dynamic range from a raw file? I obviously don't have access to the black-white gradient strips DPR uses.

By much more do you mean 10 stops? Over at the online photographer, ctien was saying he could barely eek out 10 stops from his medium format film - I doubt the ep1 can match that.

Hmm but DPR also measures over 10 stops from the ep1 using their graduated light strip: http://www.dpreview.com/reviews/olympusep1/page19.asp

Eric

DxO must be measuring something different...

No kidding. You can't use the JPEG output of a camera to determine the DR captured by the sensor, since the tone curve applied will obviously have a huge impact on the result. DxO use the RAW data.

With a proper RAW workflow you should easily get much more than 6 stops of DR from your E-P1.

--
http://picasaweb.google.com/EricSL

 

[Eric_1]

The TOP Post regarding 10 stops on MF film: http://theonlinephotographer.typepad.com/the_online_photographer/2010/02/what-cant-digital-do-for-me.html?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+typepad%2FZSjz+%28The+Online+Photographer%29&utm_content=Google+Feedfetcher
--
http://picasaweb.google.com/EricSL

 

[Ross]

I think with your method, you are just testing the in-camera EV (+ or - 3), rather than the sensor + processing rendition of the scene's brightness range. Remember DR has to do with rendering detail, not just brightness. For studio testing you could us a white terry towel and using manual exposure and find the point where the texture burns out (as in highlights) and then find the point where the texture is lost in noise. The difference in the exposure will be the DR.

Outside find a scene with a scene brightness range of 15 stops (using the spot meter in the camera), set the exposure so the histogram is at the right edge, take the picture and look at it on a monitor to see how far down into the shadows you can see detail/texture not smeared by noise. If when you metered the scene and took notes, you can figure out the DR.

I take the "who cares" position, because I don't always want to see or know what is lurking in the shadows. When I look at a picture, I like somewhere I can engage my imagination and get into the picture. In a lot of pictures shadows are a good thing for that.
--
Bob

 

[CWilt]

I aim the ep1 at a white piece of paper at iso100.

First mistake was using iso100 on the EP-1. It is fake.

DXO and DPR use different ways to measure DR and I won't get into the discussion of which one is correct. We have been down that road here before. ;-)
So you are not comparing apples to apples.

--
Charles
My family images are at http://www.stakeman.smugmug.com
Be sure of your subject.
Never, force the shot.

 

[WT21]

I was comparing the dynamic range of my ep1 with my a850 on dxo - no surprises here the a850 wins by 1.5 eV at lowest iso - when I noticed that DxO claims the ep1 has 10 stops of dynamic range!

That doesn't match with my limited testing, which suggests the ep1 has only 6 stops of dynamic range. Am I testing incorrectly? Here's my set up:

I aim the ep1 at a white piece of paper at iso100. The live histogram shows a single narrow peak indicating most pixels are at the same brightness value. Next, I increase exposure value to its max (+3), and watch that little peak shift right until its mostly in the red - this occurs at +3. Next, I decrease exposure value to its min (-3) and watch that peak shift to the left until it's almost entirely in the blue (this occurs at -3 eV)

I interpret this to mean that there are 6 stops between the brightest and darkest pixels the ep1 can simultaneously capture. Am I wrong? How on earth could the ep1 possibly get 10 full stops of dynamic range? That'd be + -5 eV of exposure compensation! DxO must be measuring something different...
--
http://picasaweb.google.com/EricSL

Just curious -- did you do the same test on your a850, and find that you can shift + - 5.5EV either way?
--

I refuse to wed myself to any of these vendors. I'm just having fun taking pictures,
and watching the technology develop.

 

[Ehrik]

I aim the ep1 at a white piece of paper at iso100.

First mistake was using iso100 on the EP-1. It is fake.

It may be better to say that it has inconsistent headroom compared to the other ISOs. It has the same DR as ISO 200 (since it's the same hardware setting) but trades one more stop of shadow range for one stop less of highlight range.

DXO and DPR use different ways to measure DR and I won't get into the discussion of which one is correct.

DXO use the engineering definition, which is max signal divided with the noise floor. The black end of this range has a signal-to-noise ratio of 1:1. This is very noisy, so most people would not accept this as a usable range. The usable range would be about two stops less.

DPR don't really measure DR, they measure the tone curve.

Any useful test of JPEG DR would have to consider the trade-off between noise and NR, smearing fine detail in the shadows. It would have to be subjective. Putting a number on it is deceptive, as it suggests some useful objective quantity has been measured.

 

[Kuivaamo]

Are you saying the live histogram counts as jpeg? How would you actually test the dynamic range from a raw file? I obviously don't have access to the black-white gradient strips DPR uses.

The live view image is processed and a tone curve is applied, so yes, it is essentially the same as a jpeg as far as DR evaluation goes.

By much more do you mean 10 stops? Over at the online photographer, ctien was saying he could barely eek out 10 stops from his medium format film - I doubt the ep1 can match that.

Not necessarily 10 stops. But definitely more than 6 and probably about 8-9 stops at base ISO, depending on what criteria you use for minimum acceptable signal over the noise floor.

 

[jkrumm]

I think dpr gets pretty close to a fair evaluation with their ACR "best" approach, looking at how much DR they can squeeze out of a camera's raw file. Runs into problems if ACR doesn't support the camera well, but this kind of test is good to me, showing real world results. So they managed 10.7 EV from the EP1, both in Olympus Studio and ACR, surprisingly (usually you don't get much out of Studio).

--
John Krumm
Juneau, AK

 

[Ehrik]

I think dpr gets pretty close to a fair evaluation with their ACR "best" approach, looking at how much DR they can squeeze out of a camera's raw file.

Problem is that how much you can recover from the point where one channel, but not all channels, clip, depends on the relative clipping points of the channels (e.g. if they clip at the same time, there is nothing to recover). This in turn depends on how the spectrum of light from the clipping subject matches the spectral responses of the colour filters in the particular camera.

This means that if camera A beats camera B on this front in the studio setting, camera B will beat camera A in some real world situations. So measuring this amount is rather pointless.

With a pointless amount added to each reported DR, the DR number itself of course gets less useful.

Then it wouldn't surprise me if the dark end clipping is also variable due to ACR treating cameras differently. The amount of sharpening etc. affects the noise magnitude (std.dev.), which can affect where the range is considered to stop. Just like DxO, this also misses the influence of pattern noise.

 

[jkrumm]

Yes, you aren't going to get any kind of exact comparison, but focusing on output is useful to users, and a reasonable approach.

--
John Krumm
Juneau, AK

 

[Andrew Westlake]

Problem is that how much you can recover from the point where one channel, but not all channels, clip, depends on the relative clipping points of the channels (e.g. if they clip at the same time, there is nothing to recover). This in turn depends on how the spectrum of light from the clipping subject matches the spectral responses of the colour filters in the particular camera.

This means that if camera A beats camera B on this front in the studio setting, camera B will beat camera A in some real world situations.

I suppose this could be true in theory, but all the Bayer cameras I've looked at show much the same behaviour in practice. Under our daylight balanced, high CRI lighting the green channel always clips first, with the blue channel roughly 2/3 stops behind and red 1/3 stop later. Do you know of any cameras which behave differently?

So measuring this amount is rather pointless.

Really? Seems to me you've identified a possible confounding factor, but you'd need concrete evidence to back it up, before you can declare what we do to be 'pointless'.

--
Andy Westlake
http://www.dpreview.com/lensreviews
http://www.dpreview.com/galleries/andy/albums

 

[CWilt]

I was trying to keep it simple for the OP.

Trading one stop highlight for the shadows being acceptable depends on the system noise.

I'm trying not to drag this into an engineering discussion for the OP.

--
Charles
My family images are at http://www.stakeman.smugmug.com
Be sure of your subject.
Never, force the shot.

 

[Ehrik]

Problem is that how much you can recover from the point where one channel, but not all channels, clip, depends on the relative clipping points of the channels (e.g. if they clip at the same time, there is nothing to recover). This in turn depends on how the spectrum of light from the clipping subject matches the spectral responses of the colour filters in the particular camera.

This means that if camera A beats camera B on this front in the studio setting, camera B will beat camera A in some real world situations.

I suppose this could be true in theory, but all the Bayer cameras I've looked at show much the same behaviour in practice. Under our daylight balanced, high CRI lighting the green channel always clips first, with the blue channel roughly 2/3 stops behind and red 1/3 stop later. Do you know of any cameras which behave differently?

Yes, DxOmark measure this. From "Analyze your camera", hoover over the "> " sign and the tab Color Response is shown. Click, and you'll get a table of "White balance scales" for the camera you've chosen.

After checking a handful of cameras, the most deviation I found was between the D300, which had

R_raw 1.56
G_raw 1
B_raw 1.28

and the A380, which had

R_raw 3.08
G_raw 1
B_raw 1.38

That's just about a full stop difference in the span between the first and last channel clipping in DxO's daylight-balanced studio lights, assuming the channels of a camera clip at the same ADU.

I've also seen spectral response curves taken from various DSLRs and they do show significant differences.

Canon, in their latest cameras, have made their red filters more transmissive to green (making them in effect yellow) in an effort to increase quantum efficiency.

So measuring this amount is rather pointless.

Really? Seems to me you've identified a possible confounding factor, but you'd need concrete evidence to back it up, before you can declare what we do to be 'pointless'.

Andy, if you think about it, even without the backing from DxO data above, what properties of the raw files, bar the different clipping, would give the differences in highlight recovery that you reviewers often find noteworthy?

 

[Andrew Westlake]

Well, if your interpretation of DxOMark data is correct, that would predict an extra stop potential highlight recovery from the A380. But Lars was only able to get the usual stop over the JPEG white point that we expect to see. Unfortunately I don't have access to the raw files to analyse directly, but on the whole I'll take Lars' word for it, based on processing both the Test shots and real-world examples, over your interpretation of DxO's test data. What matters in photography, after all, are the final images.
--
Andy Westlake
http://www.dpreview.com/lensreviews
http://www.dpreview.com/galleries/andy/albums

 

[Ehrik]

Well, if your interpretation of DxOMark data is correct

It's possible it's wrong, but I've seen spectral response curves from various DSLRs and they can vary quite a lot.

I've seen Jay Turberville's comparison of the E-330(?) and the L1, white screen in sunlight, and they were noticeably different.

His test also showed a likely imbalance in amplification between the green channels in one of the cameras, which could be due to sample variation or loose manufacturing tolerances. This could also affect R and B channels.

, that would predict an extra stop potential highlight recovery from the A380.

That's with DxOmark's studio lights and test subject . From my quick reading on the subject, colour rendering index is measured by using eight test colours, and the general CRI relates to the arithmetic mean of the Euclidean distance between chromaticities from the reference and tested light source. So there is a bit of room to play with which still gives a high CRI.

The point I was making was that there was a difference between cameras, of a magnitude easily explaining differences you would comment on in the reviews.

But Lars was only able to get the usual stop over the JPEG white point that we expect to see. Unfortunately I don't have access to the raw files to analyse directly, but on the whole I'll take Lars' word for it

I'm sure he did a honest, careful job according to the test method instructions. It's the test method itself I have a problem with.

As I noted, the light source affects things. There can be sample variation. Another possibility could be that ACR is treating cameras differently in this regard. Which again would be a rather dubious reason to comment positively or negatively on a camera .

What matters in photography, after all, are the final images.

Yes, but if your test isn't representative of real use, it becomes, as I said, rather pointless (or even misleading).

My final question in the previous post wasn't rhetorical, I'd be interested to hear your opinion; what properties of the raw files, bar the different clipping, would give the differences in highlight recovery that you reviewers often find noteworthy?

 

[Iliah Borg]

After checking a handful of cameras, the most deviation I found was between the D300, which had

R_raw 1.56
G_raw 1
B_raw 1.28

and the A380, which had

R_raw 3.08
G_raw 1
B_raw 1.38

About right. Directly in a380 arw EXIF you can see WB_RGBLevelsDaylight: 2640 1024 1696

--
http://www.libraw.org/