E-M1: Does red contrast get swallowed?

In an older thread we already noticed that red soccer shirts lost quite a bit of tonality and speculated whether this was a problem of the red channel clipping or not.

I did a comparison of the DPR studio scene with DPR's default LR settings (=no proper color profile for E-M1 yet), DPR's JPG settings and most importantly RawTherapee without any color profiles being used. My conclusion so far is that red and orange contrast seems to get swallowed by the E-M1 regardless of color profiles.

According to DxO measurements the red channel on the E-M1 separates red stronger from green than before, but the relative sensitivity of the red channel dropped. Is this a possible cause for the drop of red and orange contrast?

Here some comparisons. Sorry for switching left and right between the RawTherapee and DPR examples.


E-P5 vs. E-M1 - daylight - RAW at ISO 100 via RawTherapee (sharpened, no noise filtering, no color profile, WB picker)


E-M1 vs. E-P5 - daylight - DPR Studio Comparison Tool, RAW at ISO 200


E-M1 vs. E-P5 - daylight - DPR Studio Comparison Tool, RAW at ISO 200


E-P5 vs. E-M1 - low light - RAW at ISO 100 via RawTherapee (sharpened, no noise filtering, no color profile, WB picker)


E-M1 vs. E-P5 - low light - DPR Studio Comparison Tool, RAW at ISO 200


E-M1 vs. E-P5 - low light - DPR Studio Comparison Tool, JPG at ISO 200


E-M1 vs. E-P5 - low light - DPR Studio Comparison Tool, RAW at ISO 100


This very last JPG result may seem puzzling, but judging from other parts of the image either DPR failed to focus/shoot accurately or the JPG engine of the E-P5 doesn't work so well at ISO 100.

Just to mention it: The daylight raw file of the E-M1 at ISO 100 isn't so perfect either when compared to higher ISO files. Again either minimal misfocusing or vibrations may have happened. But the comparison of the threads vs. the E-P5 are still valid as I checked that part of the image at different ISO. Still it was reason enough for me to use nominal base ISO 200 for the DPR screenshots.

Generally I find these kind of very fine detail comparisons hard to do with the material at hand. There are too many small influences that can cause differences. My main concern at the moment is the very bad rendering of the red thread in Lightroom (hopefully remedied with the next version), which mushes reds even in the "low light" version where RawTherapee looks just fine. I am also not too much enarmored with the rendering of the red thread in the ISO 200 JPGs (not really much worse than the E-P5 JPGs, though).

Once I hold my own E-M1 in my hands I can check these things better on my finely structured red wall.


Red wall of sensor shame

This is all slightly too technical for me. However I am also finding a problem with 'reds' using Photo Ninja with the E-M1. Photo Ninja has a number of color presets you can use with any camera - scenic, portrait or neutral. Neutral doesnt turn out 'too bad' but for scenic or portrait they are extremely off.

I have contacted Photo Ninja and they confirm that their support is preliminary and that they dont have a color profile yet for the E-M1.

What I assume to be the case is that they have used a color profile of say the E-M5 as a stand in and that it is effecting the reds.

This pic below shows the problem. It is a default conversion using 'portrait' preset.

I could obviously build my own color profile for Photo Ninja (but I dont have a color checker) but in any case I suspect it will be resolved in the next release.

Yes, using a E-M5 color profile for the E-M1 does not seem advisable. In RawTherapee this causes reds to turn distinctively pinkish.

Timur Born said:

I did a comparison of the DPR studio scene with DPR's default LR settings

Click to expand...

why'd you use LR (that brings Adobe's color transforms into equation) when you have rawdigger and RPP that can do w/o any color transforms applied - so you more or less see the demosaicked image in scene-referred "color space" ("camera's color space")

Hi Timur,

Before jumping to conclusions, have a look at what I say here:

http://www.dpreview.com/forums/post/52424385

Some of the E-P5 samples may have similar problems (I haven't checked those).

The RT conversion of the artifical-light samples (what you call low light) look about equally contrasty for "red" with the E-P5 and the E-M1. This suggests that the problem with the others is due to the factor mentioned above along with over-saturation of red in the conversion color profile.

One thing you might additionally want to check (by means of RawDigger) is that the red channel on that red thread spool isn't clipped in the E-M1 RAW. I wouldn't think so, but it's something you might want to check anyway, just to be sure.

Is the effective exposure (in relation to saturation) the same? Is the light the same? Is the focusing the same?

exdeejjjaaaa said:

Timur Born said:

I did a comparison of the DPR studio scene with DPR's default LR settings

Click to expand...

why'd you use LR (that brings Adobe's color transforms into equation) when you have rawdigger and RPP that can do w/o any color transforms applied - so you more or less see the demosaicked image in scene-referred "color space" ("camera's color space")

Click to expand...

Please look again. I specifically included RawTherapee "w/o any color transforms applied", and to use RPP I'd need to use OS X, which I am not able to atm. And I also included out of camera JPGs which should use color profiles that hopefully are not just "preliminary".

Yes the "low light" samples (as called by DPR ) seem to suggest that red channel clipping (either before or after WB) may be a cause, but I am not so sure about that. Once I find time I will check via RD (need to leave now).

I also looked at the E-M5 (different lens used) and E-PL5 for comparison, just took the E-P5 as the latest example.

but also had/has 'issue' with red chanel. namely reds are very easy to and often do blow out. i do recall reading threads/reviews on this 'red' issue since the days of E1 and even before that(C5050). i haven't seen/heard much mention of the issue lately but i haven't looked/listened very hard as of late.

i wonder if other manufacturers have red chanel issues too and if the red chanel just is always the first to blow out. (issue is worse with pale caucasion skin tones which is what i shoot mainly)

--
Thanks,
Paul

Timur Born said:

Yes the "low light" samples (as called by DPR ) seem to suggest that red channel clipping (either before or after WB) may be a cause, but I am not so sure about that. Once I find time I will check via RD (need to leave now).

Click to expand...

It may additionally be a matter of too high saturation, i.e., the red being too purely red.

Timur Born said:

I also looked at the E-M5 (different lens used) and E-PL5 for comparison, just took the E-P5 as the latest example.

Click to expand...

Figured you did. ;-) And the difference is likely to be between those three (plus the E-PM2) on the one hand and the E-M1 on the other since that's the border line between different sensors.

Member said:

reds are very easy to and often do blow out.

Click to expand...

What makes you to come to this conclusion? Any raw samples with reds blown out?

Iliah Borg said:

Iliah Borg said:

reds are very easy to and often do blow out.

Click to expand...

What makes you to come to this conclusion? Any raw samples with reds blown out?

Click to expand...

he probably means OOG post color transform/wb application, not raw channel clipping.

Anders W said:

And the difference is likely

Click to expand...

Please compare http://f.cl.ly/items/3o1N1t3e052Y2a3L0U37/P9040005_ORF.txt and http://f.cl.ly/items/2t251B1J022i1u013w2o/P8130006_ORF.txt

Lightroom 5.2 colors (specifically reds) are totally borked in this release. Adobe themselves say 5.2 has "preliminary" support for the E-M1. Anyone using this release for conversions is asking for trouble. You need to use JPEG, another converter, or wait until the next release.

Iliah Borg said:

Iliah Borg said:

reds are very easy to and often do blow out.

Click to expand...

What makes you to come to this conclusion? Any raw samples with reds blown out?

Click to expand...

maybe out of gamet too in some cases...

when i did shoot predominently RAW i did not always shoot ETTR. in those cases the reds particularily in pale caucasion skin highlight areas would be completely gone/clipped/white (i don't know exactly what the reason was...just assumed clipped reds) in the jpeg file. recovery with RAW file would not always bring back much more than a flat, off-white(pink-ish/fleshy color) surface applied over clipped highlight area. no detail(s) just a flat uniformly painted area.

my bad, yes...not using tools properly.

BTW less to worry

Iliah Borg said:

Is the effective exposure (in relation to saturation) the same? Is the light the same? Is the focusing the same?

Click to expand...

These would have to be asked from DPR. They do change exposure settings and maybe light levels of their test shots by some standard unknown to me. For example the D610 is exposed considerably less than the E-M1 while using the same aperture and nominal ISO, even while the gain of the D610 is lower as measured by DxO standards.

That being said the RD histograms of the E-M1 and E-P5 shots at ISO 100 look virtually the same and the bulk of information lies within 0.1 EV when working with the images in RT. The ISO 200 are more like 0.25 EV apart, with the E-M1 shot being more exposed. In RD I placed a selection window on the red thread of the ISO 200 shots and the max values for the red channels are given as 287 (E-M1) and 256 (E-P5) respectively, out of 3839/3840 maximum. Doesn't look like a clipping issue to me then.

I do suspect that either focusing in between single shots (E-M1 daylight ISO 100 vs. 200) may either be slightly different or there has been vibrations. This is evident in the very fine line detail drawing on the left side where the ISO 100 shot misses both the vertical lines of the drawn wall and the labyrinth pattern demosaicing artifacts right below the verticals when demosaicing via RT. Originally I looked at the RAWs to examine the effect of AA filters on the various Olympus cameras and only happened to stumble across the red thread, so I noticed this one right away.

In any case, the E-P5's result on the red thread are nothing to write home about, either, at least compared to all the other colored threads. One thing I noticed is that WB is reported quite differently for both cameras when using a picker on the same target and that the orange thread of the E-M1 seems more reddish at same white patch temperature and tint. Let's wait and see what practical usage comes up with.

Timur Born said:

Iliah Borg said:

Is the effective exposure (in relation to saturation) the same? Is the light the same? Is the focusing the same?

Click to expand...

These would have to be asked from DPR.

Click to expand...

I posted the links to ISO 200 CGATS files from RawDigger above. They are created similar to http://www.rawdigger.com/usermanual/selection-grid You can use the RawDigger grid on ISO 100 files the same way. Look at WB values between the shots in the generated CGATS files, it saves a little of time.

IMO the biggest issue with reds are colour transform quality and monitor limitations.

So what's your interpretation of the data and images at hand? Normal error margin (except for the LR results)? Well possible. It's not like the RT and JPG differences are night and day.

The differences between the red thread and all the other threads are well visible, though, but I wonder now how much the 98% spectrum of the light source is responsible for that. Usually those 2% are at the extreme ends and since heat dissipation is not wanted a good deal of it likely is at the red end. No idea how spiky the spectra of those (Philips? Osram?) lights are either.

Anyway, I hoped that some people might have gathered practical experience with red contrast on the E-M1 yet and thus could contribute. I just put myself at no. 20 of a waiting list after changing preorder from Amazon to a local store, so I have to wait to see myself. The rest is rather curiosity.

Anders W said:

Timur Born said:

Yes the "low light" samples (as called by DPR ) seem to suggest that red channel clipping (either before or after WB) may be a cause, but I am not so sure about that. Once I find time I will check via RD (need to leave now).

Click to expand...

It may additionally be a matter of too high saturation, i.e., the red being too purely red.

Anders W said:

I also looked at the E-M5 (different lens used) and E-PL5 for comparison, just took the E-P5 as the latest example.

Click to expand...

Figured you did. ;-) And the difference is likely to be between those three (plus the E-PM2) on the one hand and the E-M1 on the other since that's the border line between different sensors.

Click to expand...

Also don't forget that monitors aren't able to show the whole gamut of colors, especially when colors are very saturated. So the details may be there in the raws, but it may get clipped by your monitor.