RAW Analysis: E-M5 / GH2 / GX1 / G3

[Anders W]

Using RawDigger 0.9.9 Beta at:
http://www.rawdigger.com/news/rawdigger-0-9-9

... I analyzed the following two (rated, JPG-referenced) ISO=200, F=6.3, T=1/3, EC=0 indoor E-M5 and GH2 shots that appear to have been recorded in the same lighting levels (same illuminance of the test-target scene; the same luminous-exposure in each of the camera-systems), and with each of the two cameras set to Automatic Exposure Metering modes. As good as it gets.

EM5 FW Level 1.000: http://fourthirds-user.com/sample_images/375/EM5_P3112914_ISO200.ORF

GH2 FW Level 1.0: http://fourthirds-user.com/sample_images/375/GH2_P1040143_ISO200.RW2

The E-M5 ORF is close to maximum RAW RGBG channel-levels (but not clipped in any of them).

The GH2 RW2 is slightly clipped in all RAW RGBG channels (on the order of 0.01% of photo-sites), introducing a small, but relatively insignificant error in the measurement data (relative to the E-M5).

Here are the arithmetic averages of the peak and average level statistics found (within the entire recorded image-frames, and including all of the individual RAW-level RGGB color-channels):

---- Camera -------- Peak Level -------- Average Level

------ E-M5 ------------- 3819.25 --------------- 182.95

------ GH2 -------------- 3988.50 --------------- 315.15

To determine the differences in (RAW-level) exposure between the cameras, I used this formula:

Exposure [EV] = LOG2 ( ( ( Av [EM5] ) / ( Av [GH2] ) ) / ( ( Pk [GH2] ) / ( Pk [EM5] ) ) )

Exposure [EV] = LOG2 ( ( ( 182.95 ) / ( 315.15 ) ) / ( ( 3988.50 ) / ( 3819.25 ) ) )

The numerical result is that the E-M5 exposure in this case is -0.722 EV lower than the GH2

With the reservation that I haven't tried to verify your premises or your calculations, this is fully in line with what I'd expect based on how Oly had chosen to calibrate things in the past. As already demonstrated by DxOMark, the same exposure at the same camera ISO setting yields about 2/3 EV more headroom up to the clipping point on Oly cameras compared to recent Panasonics. Compared to the oldest Panasonics, for example my G1, the difference is about 1 EV.

Potential Implications :

In the original post here (which analyzed E-M5 FW Level 1.003, and GH2 FW Level 1.0 images) at:

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40963062

... and accounting for differences in Shutter Speed of the E-M5 and GH2, the same approach derived a result where E-M5 exposure appeared to be -1.537 EV lower than GH2 exposure .

Yes, but since the shutter speed in this context cannot be used to index actual exposure (due to variable light conditions), the new results you report here (about 2/3 EV difference) are probably more reliable.

 

[Aleo Veuliah]

Aleo ,

My interest (and goal) is (as ever), to explore general technical concepts for the purpose of finding some facts (in the course of losing some illusions and gaining some knowledge). Since these DPR forums tend to be far more about "gear" than they are about actual photography or human beings, it is always my own feeling that it is the gaining and sharing of information regarding the limitations, annoyances, and flaws of (any and all) "gear" that represents a learning opportunity - whereas self-serving techno-tribal jug-fests of chest-bumping and ego-stroking are purely boring ...

I have learned some valuable things that I did not previously grasp, and hope to learn even more. Ultimately, machines are only as interesting as the human beings who create and who use them ...

DM

Well said, I also like to research and always learn new things, it is refreshing to see some threads like this one, or others showing good images or evolving photographic techniques, I also like to talk about gear but I try to not exaggerate. Other times I like to add some humor, other I am in the mood to show only images ...

All the best,

Aleo

--
Simplicity is the ultimate sophistication.

God is the tangential point between zero and infinity.

Imagination is more important than knowledge.

God always take the simplest way.

• http://aleo-photo.org

• http://blog.aleo-photo.org

 

[Franka T.L.]

I have to dig deeper into all those ...

--

• Franka -

 

[Anders W]

Given the following information regarding the test-shots existing in the DPR Studio Comp. Tool:

Andy Westlake wrote:

... you cannot deduce relative exposures between cameras from the EXIF data of these shots, because you do not know that the light levels are the same. As it happens we've recently swapped the original lighting out for a different softbox with a dimmer switch, which lets us deal better with the very high ISOs seen on the most recent cameras. We also have an ND filter gel that we sometimes use over the lights, for example with the Samsung NX200 to avoid a certain problematic shutter speed.

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40930304

Detail Man wrote:

I assume that DPReview records are kept (measured independently from model-to-model varying camera-metering readings, and measured using a separate light-meter) of the illuminance of the test target at the time of performing the individual tests.

Andy Westlake wrote:
No, they're not, as a rule. We're a camera review site, not a science lab.

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40965292

... it appears that when (apparently, from the image-file meta-data) using the same Olympus Zukio Digital ED 50mm F/2 Macro lens on both the E-M5 and the GH2, the differences in illuminance of the test-target scene when recording the E-M5 and the GH2 test-shots existing in the DPR Studio Comparison Tool was (possibly) equal to a rather significant 1.537 - 0.722 = 0.815 EV ("stops") .

This calculation rests on the presumption that the GH2 and the E-M5 were effectively given equal exposure in the DPR studio scene. However, you have previously quoted Andy saying that the E-M5 shots came out 1/3 EV darker than expected in their ISO test. If DPR has followed their own standard procedures as Andy has described them, this in turn implies that the E-M5 has been given 1/3 EV more exposure than the GH2 when shooting the studio scene samples.

If indeed the new shots that you downloaded from fourthirds-user.com have been given equal exposure (and I don't claim I know that for sure; I think I saw somebody mention in the thread about those shots that different lenses were used), you have some means to decide whether or not the same is true about the DPR shots by simply doing the same analysis with the DPR GH2 and E-M5 shots as you just did with those from the fourthirds user forum.

Note that the camera body Firmware Levels differ between the two EM5 image-files analyzed (FW Level 1.003 in the previously analyzed DPR case; FW Level 1.000 the presently analyzed FTU case). As a result of possible differences existing in JPG tone-curves and/or metering calibration, it is not possible to conclude that the difference in illuminance of the test-target scene was 0.815 EV ...

That's of course a conceivable threat to comparability. But it'd surprise me very much if they did firmware adjustments to things like those we are discussing here at this stage.

One thing that I notice (which seems perhaps rather odd, indeed), is that RawDigger reports that the Four Thirds User FW Level 1.000 E-M5 ORF RAWs have a Black Level = 0. Zero, nothing, nada .

However, if the difference in illuminance of the test-target scene was equal to 0.815 EV, that would represent a 0.408 EV difference in Signal/Noise Ratio (resulting from Photon Shot Noise sources), and one which would seem to favor the E-M5 over the GH2 in the DPR SCT JPG images.

Where it comes to the DPR SC Tool RAW images, one further complicating factor is a possibility:

Andy Westlake wrote:

Our RAW conversions are brightness-matched to the camera JPEGs so the comparison makes visual sense between cameras. This is Adobe's policy anyway - after all, it's what most users expect - so we don't generally have to adjust the output sliders from default positions, but occasionally we'll tweak the brightness control to get the best possible match.

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40965292

Adjustments of the "Brightness" control modifies the tone-curve transfer-function that is applied to the image by Camera RAW . Thus, the (visibly apparent) Signal/Noise Ratio would be changed:

The Brightness control in Camera Raw is a nonlinear adjustment that works very much like the gray input slider in Levels. It lets you redistribute the midtone values without clipping the highlights or shadows .

- Page 6

Exposure and Shadows set the white and black endpoints, respectively. Brightness adjusts the midpoint. Contrast applies an S-curve around the midpoint set by Brightness, darkening values below the midpoint and brightening those above .

- Page 5

http://www.adobe.com/digitalimag/pdfs/phscs2ip_camraw3.pdf

Being a mere mortal, I do not always get it right the first time (though I've thought carefully about the contents of this post). If a reader believes that I have missed or misinterpreted something, please feel free to present further information in a coherent and precise form that can also be considered and verified by myself as well as other readers, as the truth slowly reveals itself to us

 

[cmorse]

Using RawDigger 0.9.9 Beta at:
http://www.rawdigger.com/news/rawdigger-0-9-9

... I analyzed the following two (rated, JPG-referenced) ISO=200, F=6.3, T=1/3, EC=0 indoor E-M5 and GH2 shots that appear to have been recorded in the same lighting levels (same illuminance of the test-target scene; the same luminous-exposure in each of the camera-systems), and with each of the two cameras set to Automatic Exposure Metering modes. As good as it gets.

EM5 FW Level 1.000: http://fourthirds-user.com/sample_images/375/EM5_P3112914_ISO200.ORF

GH2 FW Level 1.0: http://fourthirds-user.com/sample_images/375/GH2_P1040143_ISO200.RW2

The E-M5 ORF is close to maximum RAW RGBG channel-levels (but not clipped in any of them).

The GH2 RW2 is slightly clipped in all RAW RGBG channels (on the order of 0.01% of photo-sites), introducing a small, but relatively insignificant error in the measurement data (relative to the E-M5).

Here are the arithmetic averages of the peak and average level statistics found (within the entire recorded image-frames, and including all of the individual RAW-level RGGB color-channels):

---- Camera -------- Peak Level -------- Average Level

------ E-M5 ------------- 3819.25 --------------- 182.95

------ GH2 -------------- 3988.50 --------------- 315.15

To determine the differences in (RAW-level) exposure between the cameras, I used this formula:

Exposure [EV] = LOG2 ( ( ( Av [EM5] ) / ( Av [GH2] ) ) / ( ( Pk [GH2] ) / ( Pk [EM5] ) ) )

Exposure [EV] = LOG2 ( ( ( 182.95 ) / ( 315.15 ) ) / ( ( 3988.50 ) / ( 3819.25 ) ) )

The numerical result is that the E-M5 exposure in this case is -0.722 EV lower than the GH2

Bringing them into RawTherapee if all the auto settings stripped away the -.722 EV looks about right when comparing the paper. Pushing the shadows (the few there are) it looks like they have roughly the same amount of detail with the GH2 having more noise. In the highlights on the brass the gh2 are completely clipped. On the EM5 there is little to no clipping on the brass. You can actually see that the light source is a row of three window and can faintly see details of outside in the reflection.

Over all, given the clipping and shadow detail, of the .722 difference I'd probably say the GH2 is .5EV over exposed and the EM5 is about .25EV under exposed.

 

[Detail Man]

Detail Man wrote:

... it appears that when (apparently, from the image-file meta-data) using the same Olympus Zukio Digital ED 50mm F/2 Macro lens on both the E-M5 and the GH2, the differences in illuminance of the test-target scene when recording the E-M5 and the GH2 test-shots existing in the DPR Studio Comparison Tool was (possibly) equal to a rather significant 1.537 - 0.722 = 0.815 EV ("stops") .

This calculation rests on the presumption that the GH2 and the E-M5 were effectively given equal exposure in the DPR studio scene.

The (original post) 1.537 EV derived from the calculation involving analysis of the DPR studio-scene (where the same lens appears to have been used) normalizes for the difference in Shutter Speeds, and is not presented (as quoted, above) on a premise that the exposures (involving studio-scene illuminance, as well as presumed equal T-Numbers, and the normalized Shutter Speeds) are equal.

If indeed the new shots that you downloaded from fourthirds-user.com have been given equal exposure (and I don't claim I know that for sure; I think I saw somebody mention in the thread about those shots that different lenses were used), ...

You make a valid point. The image-file meta-data of the fourthirds-user.com shots themselves report that an Olympus 12-50mm lens was used on the E-M5, and a Lumix G Vario 14-140mm lens was used on the GH2. Due to possible differences in the lens-transmission-factors, the T-Numbers may differ somewhat. In order for the difference calculation above [ 1.537 - 0.722 = 0.815 EV ("stops") ] to be (absolutely) valid, it is necessary that the exposure be equal between the fourthirds-user.com shots. The difference in the T-Numbers between the two lenses used is an uncertainty. Thus, it can't be said that the fourthirds-user.com shots have exactly equal exposure.

Note that the camera body Firmware Levels differ between the two EM5 image-files analyzed (FW Level 1.003 in the previously analyzed DPR case; FW Level 1.000 the presently analyzed FTU case). As a result of possible differences existing in JPG tone-curves and/or metering calibration, it is not possible to conclude that the difference in illuminance of the test-target scene was 0.815 EV ...

That's of course a conceivable threat to comparability. But it'd surprise me very much if they did firmware adjustments to things like those we are discussing here at this stage.

I do not know. I accept your speculation about that - but it only seemed reasonable to note the differences in FW-Level as a (possible) unknown affecting the E-M5's JPG tone-curves and/or metering calibration.

Based upon a premise that the change in FW-Level did not change the E-M5's JPG tone-curves and/or metering calibration, it could be said that (some) of those effects (namely, the metering calibration) would be common to the results derived from both calculations.

However , it is important to here note (as I have not previously grasped, or expressed) that since the RGB tone-curve transfer-functions are non-linear in nature, and since the (peak as well as average) signal-levels between the DPR and FTU shots are not identical, there does indeed exist an (essentially unknowable to us) difference in tone-curve transfer-function scaling-factor that does indeed seem to inject a further uncertainty into the EV difference calculation that I have made ...

My intention in making the difference calculation above [ 1.537 - 0.722 = 0.815 EV ("stops") ] was to (on a presumption that the fourthirds-user.com shots have nearly equal exposure - with your point made regarding the different lenses at the same F-Number acknowledged and accepted) derive a difference in EV that would reflect the differences in illuminance of the DPR studio-scene when the DPR test-shots were recorded.

However, I do now (at this point) believe that the above mentioned (and essentially unknowable) differences in the non-linear tone-curve transfer-functions (for each individual camera, and at difference peak and average signal-levels as they exist in the DPR and the FTU test-shots) introduce an additional factor that does not allow the originally intended isolation of the differences in illuminance of the DPR studio-scene. Thank you for causing me to think more/harder about this.

However, you have previously quoted Andy saying that the E-M5 shots came out 1/3 EV darker than expected in their ISO test. If DPR has followed their own standard procedures as Andy has described them, this in turn implies that the E-M5 has been given 1/3 EV more exposure than the GH2 when shooting the studio scene samples.

I don't recall quoting them (perhaps I have), but I am indeed aware of Andy Westlake's statements that you refer to (directly above). I have no reason to doubt Andy's sincerity, or integrity of memory. My goal has been to (try) to derive some meaningful possibilities from the data that is available (with all of it's unknowns, not the least of them being unrecorded DPR illumination-levels), in an attempt to estimate differences in exposure affecting Signal/Noise Ratios.

If indeed the new shots that you downloaded from fourthirds-user.com have been given equal exposure (and I don't claim I know that for sure; I think I saw somebody mention in the thread about those shots that different lenses were used), you have some means to decide whether or not the same is true about the DPR shots by simply doing the same analysis with the DPR GH2 and E-M5 shots as you just did with those from the fourthirds user forum.

Given the levels of complexity/uncertainty that our conversation has revealed, that seems unclear ?

 

[Detail Man]

Anders W wrote:

If indeed the new shots that you downloaded from fourthirds-user.com have been given equal exposure (and I don't claim I know that for sure; I think I saw somebody mention in the thread about those shots that different lenses were used), you have some means to decide whether or not the same is true about the DPR shots by simply doing the same analysis with the DPR GH2 and E-M5 shots as you just did with those from the fourthirds user forum.

Given the levels of complexity/uncertainty that our conversation has revealed, that seems unclear ?

More specifically, the formula used in the original post:

At rated ISO=200, these are the numerical RAW-level clipping values [the maximum values shown in parentheses below represent the (actual) maximum after dark-noise subtraction takes place]. Also listed below are the Shutter Speeds that DPReview testers used for each individual camera:

Camera -------- Maximum RAW Level -------- Shutter Speed

-- E-M5 ---------------- 4049 (3796) ------------------------- 50
-- GH2 ----------------- 3989 (3974) ------------------------- 80

Using that data that RawDigger finds for the E-M5, in order to determine the difference in EV of the E-M5 relative to each of the other cameras' light-metering systems, this is the correct formula:

Difference [in EV] = LOG2 ( ( (AVG) / (1096.2) ) x ( (3796) / (MRL) ) x ( (S) / (50) ) )

where:

AVG is the arithmetic sum of all of the individual RGGB Channel Averages shown in RawDigger (here are the values that you will get: GH2=2081.8; GX1=2011.0; G3=1738.5; and E-PL3=1441.9);

MRL is the Maximum RAW Level ( after dark-noise subtraction ) for each camera (see table above);

S is the value of Shutter Speed used by each camera in the DPR test-shot (see table above).
.

Results [ the number of EV that the E-M5 under-exposes by relative to other cameras ]:

Relative to the GH2: 1.537 EV

... is different in nature from the formula used in my more recent post:

Here are the arithmetic averages of the peak and average level statistics found (within the entire recorded image-frames, and including all of the individual RAW-level RGGB color-channels):

---- Camera -------- Peak Level -------- Average Level

------ E-M5 ------------- 3819.25 --------------- 182.95

------ GH2 -------------- 3988.50 --------------- 315.15

To determine the differences in (RAW-level) exposure between the cameras, I used this formula:

Exposure [EV] = LOG2 ( ( ( Av [EM5] ) / ( Av [GH2] ) ) / ( ( Pk [GH2] ) / ( Pk [EM5] ) ) )

Exposure [EV] = LOG2 ( ( ( 182.95 ) / ( 315.15 ) ) / ( ( 3988.50 ) / ( 3819.25 ) ) )

The numerical result is that the E-M5 exposure in this case is -0.722 EV lower than the GH2

... only in the sense that the (in the DPR case, non-equal) Shutter Speeds were factored-in (in the case of the first formula), and (in the second formula) the reciprocal form was stated (so that the result in EV would have a negative, as oppose to a positive, sign in the numerical result).

Due to DPR's statement that the illuminance of the studio-scenes is not necessarily equal, since (in the DPR case) the same lens was used, and the differences in Shutter Speed were factored in, the 0.815 EV difference between the results of the two calculations may tend to imply that the exposure (a function of illuminance, T-Number, and Shutter Speed) of the E-M5 and GH2 in the DPR test-shots was not equal. However, it does indeed appear that there's an additional unknown.

As I have mentioned, the fact that the peak and average signal-levels differ between the DPR and the FTU test-shots, and the effects of the non-linear nature of the RGB tone-curve transfer-functions of both cameras upon their (JPG-referenced) metering further "clouds the waters" ...

 

[Detail Man]

One thing that I notice (which seems perhaps rather odd, indeed), is that RawDigger reports that the Four Thirds User FW Level 1.000 E-M5 ORF RAWs have a Black Level = 0. Zero, nothing, nada .

I do not know quite what to make of this behavior of the EM-5 FW Level 1.000. As a result of that behavior, I made the calculations below without subtracting black-level in the source statistics.

Using RawDigger 0.9.9 Beta at:
http://www.rawdigger.com/news/rawdigger-0-9-9

... I analyzed the following two (rated, JPG-referenced) ISO=200, F=6.3, T=1/3, EC=0 indoor E-M5 and GH2 shots that appear to have been recorded in the same lighting levels (same illuminance of the test-target scene; the same luminous-exposure in each of the camera-systems), and with each of the two cameras set to Automatic Exposure Metering modes. As good as it gets.

EM5 FW Level 1.000: http://fourthirds-user.com/sample_images/375/EM5_P3112914_ISO200.ORF

GH2 FW Level 1.0: http://fourthirds-user.com/sample_images/375/GH2_P1040143_ISO200.RW2

The E-M5 ORF is close to maximum RAW RGBG channel-levels (but not clipped in any of them).

The GH2 RW2 is slightly clipped in all RAW RGBG channels (on the order of 0.01% of photo-sites), introducing a small, but relatively insignificant error in the measurement data (relative to the E-M5).

Here are the arithmetic averages of the peak and average level statistics found (within the entire recorded image-frames, and including all of the individual RAW-level RGGB color-channels):

---- Camera -------- Peak Level -------- Average Level

------ E-M5 ------------- 3819.25 --------------- 182.95

------ GH2 -------------- 3988.50 --------------- 315.15

To determine the differences in (RAW-level) exposure between the cameras, I used this formula:

Exposure [EV] = LOG2 ( ( ( Av [EM5] ) / ( Av [GH2] ) ) / ( ( Pk [GH2] ) / ( Pk [EM5] ) ) )

Exposure [EV] = LOG2 ( ( ( 182.95 ) / ( 315.15 ) ) / ( ( 3988.50 ) / ( 3819.25 ) ) )

The numerical result is that the E-M5 exposure in this case is -0.722 EV lower than the GH2
.

Potential Implications :

In the original post here (which analyzed E-M5 FW Level 1.003, and GH2 FW Level 1.0 images) at:

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40963062

... and accounting for differences in Shutter Speed of the E-M5 and GH2, the same approach derived a result where E-M5 exposure appeared to be -1.537 EV lower than GH2 exposure .

... when (apparently, from the image-file meta-data) using the same Olympus Zukio Digital ED 50mm F/2 Macro lens on both the E-M5 and the GH2, the differences in illuminance of the test-target scene when recording the E-M5 and the GH2 test-shots existing in the DPR Studio Comparison Tool was (possibly) equal to a rather significant 1.537 - 0.722 = 0.815 EV ("stops").

 

[Detail Man]

Below is an alternative calculation conducted with the GH2 black-level (of 15) subtracted from the GH2 peak and average RAW-level statistics (in a situation where RawDigger reports no E-M5 black-level to be subtracted), which is consistent with the method found in my original post.

Here are the arithmetic averages of the peak and average level statistics found (within the entire recorded image-frames, and including all of the individual RAW-level RGGB color-channels):

---- Camera -------- Peak Level -------- Average Level

------ E-M5 ------------- 3819.25 --------------- 182.95

------ GH2 ------------- 3973.50 -------------- 300.15

To determine the differences in (RAW-level) exposure between the cameras, I used this formula:

Exposure [EV] = LOG2 ( ( ( Av [EM5] ) / ( Av [GH2] ) ) / ( ( Pk [GH2] ) / ( Pk [EM5] ) ) )

Exposure [EV] = LOG2 ( ( ( 182.95 ) / ( 300.15 ) ) / ( ( 3973.50 ) / ( 3819.25 ) ) )

The numerical result is that the E-M5 exposure in this case is -0.657 EV lower than the GH2
.

In the original post here (which analyzed E-M5 FW Level 1.003, and GH2 FW Level 1.0 images) at:

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40963062

... and accounting for differences in Shutter Speed of the E-M5 and GH2, the same approach derived a result where E-M5 exposure appeared to be -1.537 EV lower than GH2 exposure .

... when (apparently, from the image-file meta-data) using the same Olympus Zukio Digital ED 50mm F/2 Macro lens on both the E-M5 and the GH2, the differences in illuminance of the test-target scene when recording the E-M5 and the GH2 test-shots existing in the DPR Studio Comparison Tool was (possibly) equal to a rather significant 1.537 - 0.657 = 0.880 EV ("stops").

Please note that in posts subsequent to the above (quoted, and modified) text, I have discussed further unknown factors which would seem to (also) be a factor - with the effect of rendering the above derived 0.880 EV difference in EVs between the two calculations as not constituting a reliable estimation of differences in illuminance of the test-target scene . I have (nevertheless) included that (modified) text in this (addendum) post including modified, alternative calculations.
.

The (alternative calculation's derived) -0.657 EV differences in (RAW-level) exposure between the E-M5 and GH2 is quite close to the approximately -2/3 EV ("stop") difference that Anders W appears to refer to, as well as the approximately -2/3 EV ("stop") RAW-level difference that Iliah Borg has reported as existing between the Panasonic and Olympus camera models he has tested.

With respect to Signal/Noise Ratio resulting from Photon Shot Noise, this information (in the case of all other factors being equal) would appear to place the E-M5 at a 1/3 EV ("stop") disadvantage relative to the GH2 (if/when the users were to rely upon automatic in-camera metering indications)

Please note that the above statements relate to RAW-mode (not JPG-mode-referenced) levels - but a 1/3 EV ("stop") disadvantage in Signal/Noise Ratio would impact E-M5 JPGs as well as RAWs .

DM ...

 

[Anders W]

Detail Man wrote:

... it appears that when (apparently, from the image-file meta-data) using the same Olympus Zukio Digital ED 50mm F/2 Macro lens on both the E-M5 and the GH2, the differences in illuminance of the test-target scene when recording the E-M5 and the GH2 test-shots existing in the DPR Studio Comparison Tool was (possibly) equal to a rather significant 1.537 - 0.722 = 0.815 EV ("stops") .

This calculation rests on the presumption that the GH2 and the E-M5 were effectively given equal exposure in the DPR studio scene.

The (original post) 1.537 EV derived from the calculation involving analysis of the DPR studio-scene (where the same lens appears to have been used) normalizes for the difference in Shutter Speeds, and is not presented (as quoted, above) on a premise that the exposures (involving studio-scene illuminance, as well as presumed equal T-Numbers, and the normalized Shutter Speeds) are equal.

I think we misunderstood each other here. What I mean by "equal exposure" is "same amount of light", which if the light level varies between cameras (as seems to be the case for the DPR studio scene) of course implies different shutter speeds. Further, I didn't say that you presented your conclusion as resting on the presumption I mentioned. Rather, I pointed out that as a matter of fact it does. Unless you make that presumption, your deduction that the difference in illuminance amounts to 0.815 EV cannot be made.

I do not know. I accept your speculation about that - but it only seemed reasonable to note the differences in FW-Level as a (possible) unknown affecting the E-M5's JPG tone-curves and/or metering calibration.

Yes, you certainly did right to note the difference and mention the possibility.

However, you have previously quoted Andy saying that the E-M5 shots came out 1/3 EV darker than expected in their ISO test. If DPR has followed their own standard procedures as Andy has described them, this in turn implies that the E-M5 has been given 1/3 EV more exposure than the GH2 when shooting the studio scene samples.

I don't recall quoting them (perhaps I have), but I am indeed aware of Andy Westlake's statements that you refer to (directly above). I have no reason to doubt Andy's sincerity, or integrity of memory. My goal has been to (try) to derive some meaningful possibilities from the data that is available (with all of it's unknowns, not the least of them being unrecorded DPR illumination-levels), in an attempt to estimate differences in exposure affecting Signal/Noise Ratios.

The post I had in mind when saying that you quoted him is this:

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40965303

If indeed the new shots that you downloaded from fourthirds-user.com have been given equal exposure (and I don't claim I know that for sure; I think I saw somebody mention in the thread about those shots that different lenses were used), you have some means to decide whether or not the same is true about the DPR shots by simply doing the same analysis with the DPR GH2 and E-M5 shots as you just did with those from the fourthirds user forum.

Given the levels of complexity/uncertainty that our conversation has revealed, that seems unclear ?

What I meant is the following: If we accept, as a working hypothesis, that the shots from the fourthirds-user forum were indeed given equal exposure (the same amount of light), then you can check whether or not that is true for the DPR shots as well by simply performing exactly the same analysis on them as you just did on the shots from fourthirds-user. If the cameras were given the same amount of light in both cases, then your analysis should give roughly the same results in both cases. Do you follow me?

 

[Detail Man]

... If we accept, as a working hypothesis, that the shots from the fourthirds-user forum were indeed given equal exposure (the same amount of light), then you can check whether or not that is true for the DPR shots as well by simply performing exactly the same analysis on them as you just did on the shots from fourthirds-user. If the cameras were given the same amount of light in both cases, then your analysis should give roughly the same results in both cases. Do you follow me?

Yes, I do, and I agree. But the fact is that the results are different - which is what caused me to consider that the illuminance of the DPR test-scene was different. That was the point of my posts:

Potential Implications :

In the original post here (which analyzed E-M5 FW Level 1.003, and GH2 FW Level 1.0 images) at:

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40963062

... and accounting for differences in Shutter Speed of the E-M5 and GH2, the same approach derived a result where E-M5 exposure appeared to be -1.537 EV lower than GH2 exposure .

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40994408

... it appears that when (apparently, from the image-file meta-data) using the same Olympus Zukio Digital ED 50mm F/2 Macro lens on both the E-M5 and the GH2, the differences in illuminance of the test-target scene when recording the E-M5 and the GH2 test-shots existing in the DPR Studio Comparison Tool was (possibly) equal to a rather significant 1.537 - 0.722 = 0.815 EV ("stops") .

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40995281

Note that I subsequently revised my viewpoint do to the existence of the following factor:

However , it is important to here note (as I have not previously grasped, or expressed) that since the RGB tone-curve transfer-functions are non-linear in nature, and since the (peak as well as average) signal-levels between the DPR and FTU shots are not identical, there does indeed exist an (essentially unknowable to us) difference in tone-curve transfer-function scaling-factor that does indeed seem to inject a further uncertainty into the EV difference calculation that I have made ...

My intention in making the difference calculation above [ 1.537 - 0.722 = 0.815 EV ("stops") ] was to (on a presumption that the fourthirds-user.com shots have nearly equal exposure - with your point made regarding the different lenses at the same F-Number acknowledged and accepted) derive a difference in EV that would reflect the differences in illuminance of the DPR studio-scene when the DPR test-shots were recorded.

However, I do now (at this point) believe that the above mentioned (and essentially unknowable) differences in the non-linear tone-curve transfer-functions (for each individual camera, and at difference peak and average signal-levels as they exist in the DPR and the FTU test-shots) introduce an additional factor that does not allow the originally intended isolation of the differences in illuminance of the DPR studio-scene. Thank you for causing me to think more/harder about this.

http://forums.dpreview.com/forums/read.asp?forum=1041&message=41002262

Note that (above) I also recognized your point made the (in the case of the FTU test-shots), the lenses used were different (and thus may have had a different lens-transmission factor, causing different T-Numbers even though the F-Numbers were the same). Thus, neither of the two pairs of test shots (the FTU or the DPR) can be said to be directly comparable, due (in part) to the fact that the FTU test-shots cannot be said to have the same "exposure" (as you have referred to it's accepted definition here):

Anders W wrote:

... a certain exposure (the amount of light the sensor is given as determined by the T-stop, the shutter speed, and the light conditions).

http://forums.dpreview.com/forums/read.asp?forum=1041&message=41008653

Anders W wrote:

This calculation rests on the presumption that the GH2 and the E-M5 were effectively given equal exposure in the DPR studio scene.

Actually, the calculation rested on the presumption that the GH2 and E-M5 were effectively given the same exposure in the FTU test-shots (which your pointing out that different lenses were used in the FTU test-shots has rightfully called into question), and the very purpose of performing the calculation was (with the differences in the Shutter Speeds of the GH2 and the E-M5 accounted for, and with the same lens used on both cameras in that case) to attempt to derive the amount that the illuminance of the DPR studio scene varied between the GH2 and the E-M5 test-shots ...

... What I mean by "equal exposure" is "same amount of light", which if the light level varies between cameras (as seems to be the case for the DPR studio scene) of course implies different shutter speeds. Further, I didn't say that you presented your conclusion as resting on the presumption I mentioned. Rather, I pointed out that as a matter of fact it does. Unless you make that presumption, your deduction that the difference in illuminance amounts to 0.815 EV cannot be made.

... so your statement (directly above) still confuses me, and seems contradictory. However, it has been a series of long and involved statements made between us, and there has likely somewhere evolved some sort of misunderstanding in terms. I suspect that we would ultimately agree, and it is all rendered somewhat moot by the potential difference in FTU test-shot T-Numbers, anyway ..

 

[Kuppenbender]

This is not surprising. In fact it only makes sense. Back when MP3 was first invented, the audiophiles where screaming in agony that the format simply clipped away the frequencies that the human ear cannot hear. Blasphemy!

No screaming, no agony, simply an explanation as to the difference in sound. If you mainly listen to music on your car radio or an iPod dock (nothing wrong with that) you're absoloutely right. No discernible difference. Whatever floats your boat.

 

[Detail Man]

This is not surprising. In fact it only makes sense. Back when MP3 was first invented, the audiophiles where screaming in agony that the format simply clipped away the frequencies that the human ear cannot hear. Blasphemy!

No screaming, no agony, simply an explanation as to the difference in sound. If you mainly listen to music on your car radio or an iPod dock (nothing wrong with that) you're absoloutely right. No discernible difference. Whatever floats your boat.

I, myself, personally like my 96 dB CD DR just fine, and losing around 16 DB of DR to schemes that slice/dice into the upper-frequency bands degrade wide-band spectral sources noticeably to my ear - unless upper frequency response of the reproduction system is already limited by other roll-offs ...

 

[Detail Man]

Was able to locate some lens-transmission factor data for the two lenses involved (as indicated by the image-file meta-data), and formulate what is a pretty good estimate of effects upon exposure:

Olympus M.ZUIKO DIGITAL ED 12-50mm f3.5-6.3 EZ:

http://www.dxomark.com/index.php/Lenses/Camera-Lens-Database/Olympus/Olympus-MZUIKO-DIGITAL-ED-12-50mm-F35-63-EZ/%28camera%29/677

Panasonic LUMIX G Vario HD 14-140mm F4.0-5.8 ASPH:

http://www.dxomark.com/index.php/Lenses/Camera-Lens-Database/Panasonic/Panasonic-LUMIX-G-Vario-HD-14-140mm-F40-58-ASPH

The difference in T-Stops around the commonly set F=6.3 appears to be around 0.15 EV (with more light-loss in the GH2's LGV 14-140mm than the E-M5's Oly 12-50mm). Adjustments below:

Below is an alternative calculation conducted with the GH2 black-level (of 15) subtracted from the GH2 peak and average RAW-level statistics (in a situation where RawDigger reports no E-M5 black-level to be subtracted), which is consistent with the method found in my original post.

Curiously, but true, the particular (FW Level 1.000) E-M5 image-file recorded by 4/3 User that was analyzed is reported by RawDigger as having a Zero(0) Black Level. This is different than the DPR (FW Level 1.003) Studio Comparison Tool RAWs (which shows a Black Level in the range of 250 - 254). Not sure what to make of that difference (which appears to affect the DPR test-shots) ...

Here are the arithmetic averages of the peak and average level statistics found (within the entire recorded image-frames, and including all of the individual RAW-level RGGB color-channels):

---- Camera -------- Peak Level -------- Average Level

------ E-M5 ------------- 3819.25 --------------- 182.95

------ GH2 ------------- 3973.50 -------------- 300.15

To determine the differences in (RAW-level) exposure between the cameras, I used this formula:

Exposure [EV] = LOG2 ( ( ( Av [EM5] ) / ( Av [GH2] ) ) / ( ( Pk [GH2] ) / ( Pk [EM5] ) ) )

Exposure [EV] = LOG2 ( ( ( 182.95 ) / ( 300.15 ) ) / ( ( 3973.50 ) / ( 3819.25 ) ) )

The numerical result is that the E-M5 exposure in this case is -0.507 EV lower than the GH2

In the original post here (which analyzed E-M5 FW Level 1.003, and GH2 FW Level 1.0 images) at:

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40963062

... and accounting for differences in Shutter Speed of the E-M5 and GH2, the same approach derived a result where E-M5 exposure appeared to be -1.537 EV lower than GH2 exposure .

... when (apparently, from the image-file meta-data) using the same Olympus Zukio Digital ED 50mm F/2 Macro lens on both the E-M5 and the GH2, the differences in illuminance of the test-target scene when recording the E-M5 and the GH2 test-shots existing in the DPR Studio Comparison Tool was ( possibly ) equal to a rather significant 1.537 - 0.507 = 1.03 EV ("stops").

Please note that in posts subsequent to the above (quoted, and modified) text, I have discussed further unknown factors which would seem to (also) be a factor - with the effect of rendering the above derived 1.03 EV difference in EVs between the two calculations as not constituting a reliable estimation of differences in illuminance of the test-target scene. I have (nevertheless) included that (modified) text in this (addendum) post including modified, alternative calculations .

With respect to Signal/Noise Ratio resulting from Photon Shot Noise, this information (in the case of all other factors being equal) would appear to place the E-M5 at a 1/4 EV ("stop") disadvantage relative to the GH2 (if/when the users were to rely upon automatic in-camera metering indications)

Please note that the above statements relate to RAW-mode (not JPG-mode-referenced) levels - but a 1/4 EV ("stop") disadvantage in Signal/Noise Ratio would impact E-M5 JPGs as well as RAWs .

 

[Detail Man]

Was able to locate some lens-transmission factor data for the two lenses involved (as indicated by the image-file meta-data), and formulate what is a pretty good estimate of effects upon exposure:

Olympus M.ZUIKO DIGITAL ED 12-50mm f3.5-6.3 EZ:

http://www.dxomark.com/index.php/Lenses/Camera-Lens-Database/Olympus/Olympus-MZUIKO-DIGITAL-ED-12-50mm-F35-63-EZ/%28camera%29/677

Panasonic LUMIX G Vario HD 14-140mm F4.0-5.8 ASPH:

http://www.dxomark.com/index.php/Lenses/Camera-Lens-Database/Panasonic/Panasonic-LUMIX-G-Vario-HD-14-140mm-F40-58-ASPH

The difference in T-Stops around the commonly set F=6.3 appears to be around 0.15 EV (with more light-loss in the GH2's LGV 14-140mm than the E-M5's Oly 12-50mm). Adjustments below:

Below is an alternative calculation conducted with the GH2 black-level (of 15) subtracted from the GH2 peak and average RAW-level statistics (in a situation where RawDigger reports no E-M5 black-level to be subtracted), which is consistent with the method found in my original post.

Curiously, but true, the particular (FW Level 1.000) E-M5 image-file recorded by 4/3 User that was analyzed is reported by RawDigger as having a Zero(0) Black Level. This is different than the DPR (FW Level 1.003) Studio Comparison Tool RAWs (which shows a Black Level in the range of 250 - 254). Not sure what to make of that difference (which appears to affect the DPR test-shots) ...

Here are the arithmetic averages of the peak and average level statistics found (within the entire recorded image-frames, and including all of the individual RAW-level RGGB color-channels):

---- Camera -------- Peak Level -------- Average Level

------ E-M5 ------------- 3819.25 --------------- 182.95

------ GH2 ------------- 3973.50 -------------- 300.15

To determine the differences in (RAW-level) exposure between the cameras, I used this formula:

Exposure [EV] = LOG2 ( ( ( Av [EM5] ) / ( Av [GH2] ) ) / ( ( Pk [GH2] ) / ( Pk [EM5] ) ) )

Exposure [EV] = LOG2 ( ( ( 182.95 ) / ( 300.15 ) ) / ( ( 3973.50 ) / ( 3819.25 ) ) )

The numerical result is that the E-M5 exposure in this case is -0.507 EV lower than the GH2

In the original post here (which analyzed E-M5 FW Level 1.003, and GH2 FW Level 1.0 images) at:

http://forums.dpreview.com/forums/read.asp?forum=1041&message=40963062

... and accounting for differences in Shutter Speed of the E-M5 and GH2, the same approach derived a result where E-M5 exposure appeared to be -1.537 EV lower than GH2 exposure .

... when (apparently, from the image-file meta-data) using the same Olympus Zukio Digital ED 50mm F/2 Macro lens on both the E-M5 and the GH2, the differences in illuminance of the test-target scene when recording the E-M5 and the GH2 test-shots existing in the DPR Studio Comparison Tool was ( possibly ) equal to a rather significant 1.537 - 0.507 = 1.03 EV ("stops").

Please note that in posts subsequent to the above (quoted, and modified) text, I have discussed further unknown factors which would seem to (also) be a factor - with the effect of rendering the above derived 1.03 EV difference in EVs between the two calculations as not constituting a reliable estimation of differences in illuminance of the test-target scene. I have (nevertheless) included that (modified) text in this (addendum) post including modified, alternative calculations .

With respect to Signal/Noise Ratio resulting from Photon Shot Noise, this information (in the case of all other factors being equal) would appear to place the E-M5 at a 1/4 EV ("stop") disadvantage relative to the GH2 (if/when the users were to rely upon automatic in-camera metering indications)

Please note that the above statements relate to RAW-mode (not JPG-mode-referenced) levels - but a 1/4 EV ("stop") disadvantage in Signal/Noise Ratio would impact E-M5 JPGs as well as RAWs .

Have discovered that one of the first E-M5 ORFs recorded by Ian at M4/3 User ("EM5-2.ORF"; the indoor shots of the racks of jackets in the clothing store) was shot with (I presume) the same (FW Level 1.000) E-M5 (ISO=200, F=6.3, T=1/6), but reports a Black Level of 253, definitely clipping in all RAW-channels at 3796, the sum of those two numbers equaling 4097 .

Yet, the later test-shot with (I presume) the same (FW Level 1.000) E-M5 ("P3112914.ORF", ISO=200, F=6.3, T=1/3 ) reports a Black Level of Zero (0), (also) definitely clipping in all RAW-channels at 3820 . With an absolute maximum RAW-level of (the same) 4097, this would imply a Black Level of 229 ... ... I do not know just what to make of that discrepancy ...

At any rate, the above quoted analysis remains consistent (using whatever Black Level actually existed in each of the RAW files compared) - and it appears that a Black Level of 229 does in fact exist in that E-M5 image-file (even though for some unknown reason RawDigger does not find it).

DM ...