Sony ARW looks yellow in RT

pgoelz · December 14, 2019, 5:30pm

I’m trying to get on good terms with RT in the event I decide that CaptureOne’s pricing model is excessive. As a test, I have been playing with some images that I have also processed in CaptureOne to see what kind of results I can get. For the most part, RT seems to be able to get similar results and I like the price I also really like the fact that RT does not use a database (catalog).

However, I am unable to get an acceptable white balance in RT on images that look perfect in CaptureOne without any modification. No matter what I do with the color temperature, tint and even channel mixer sliders, the resulting image looks excessively yellow… or perhaps just insufficient red? Even using the dropper on a white area does not produce an acceptable white balance… it always looks just a bit yellow. Like maybe the curves on the three colors are not the same? If so, I have not found where you can change that.

I have tried all the bundled profiles and all the various de-mosaicing settings with the same result.

Have I missed something? I can post examples if it will help.

Paul

paperdigits · December 14, 2019, 5:40pm

Yes, please do. If you can post the raw file, others will process it and share their sidecar file so you can inspect their results.

pgoelz · December 14, 2019, 8:01pm

OK, here is an example.

This is just a very noisy test shot in the family room at max ISO. I’m using it as an example because the yellow tint is quite noticeable.

The RAW:
DSC04493.ARW (24.1 MB)

The CaptureOne version (minimal adjustments, no color temperature manipulation):

The RT version:

The *.pp3 file:
DSC04493.ARW.pp3 (11.9 KB)

And a second example. On first glance they appear similar but I find a more pleasing depth of color in the CaptureOne version, especially in the shadows like in her forearms.

The RAW:
DSC04425.ARW (23.8 MB)

The CaptureOne version:

Two RT versions:

The *.pp3 file:
DSC04425.ARW.pp3 (11.9 KB)

Overall, I find that the RawTherapee images are yellowish in spite of my best efforts with the color temperature, black point, shadow compression, saturation and color mixer sliders.

???

Paul

heckflosse · December 14, 2019, 9:13pm

My RT version. I did not care about the colors, but about sharpening. Clearly less sharpening halos in my RT version compared to your CO version

ggbutcher · December 14, 2019, 9:28pm

I messed with the bug picture a bit in both RT and my hack software; the upshot is I think you have a lot going on in the renditions you posted, and you need to tease it all apart.

In RT, start with the Neutral profile. When I did that, I found it curious that the image was definitely more yellowish than my hack software’s neutral, which uses the same camera primaries as RT, and the camera white balance, which was what the RT Neutral profile also uses. For the camera whitebalance, RT showed temp/tint of 4750/1.0; I was able to approximate your target color with 3760/0.810 To get 3760/0.810, I started with dragging temp away from yellow (makes sense to me… ), but I found it going too blue before I got to pleasing green. So, I drug tint away from that cyan dot and found joy…

I’m more used to dealing with the camera RGB multipliers, so I’m going to dig a bit to understand the RT conversion of those to temp/tint.

Now, adjusting color at Neutral will be a bit counter-intuitive, as the overall image is still dull. After you get color, next will be to adjust tone, making parts brighter/darker, adding/subtracting contrast, etc. using the rich selection of tone operators. The tone operators will to varying degrees mess with color, so you need to get color where you want it first…

Soooo… start with Neutral, adjust color, then adjust tone. For what it’s worth…

WRT RT’s presentation of a warmer starting color, I’ll have to let one of the RT practitioners weigh in… to me, it looks like some sort of difference in interpretation of the camera white balance.

gaaned92 · December 15, 2019, 8:33am

I observe that the preview is different from the generated jpg

prview:

generated jpg:

xfer · December 15, 2019, 11:53am

Looks like a difference in camera input color profile, to me.
Capture One color profiles are notoriously finely-tuned to have the best results; RT profiles are good, but not always to CO level.
If the OP has the time/interest, he could try building a custom color profile for his camera (with XRite ColorChecker Passport, for example). Not the easiest task, but usually worth the effort.

I don’t know if it’s possible to use CO profiles in RT; maybe different RGB multipliers etc.

gwaugh · December 15, 2019, 1:17pm

My quick attempt. Looks over exposed to me.

DSC04493.ARW.pp3 (11.3 KB)

gwaugh · December 15, 2019, 1:21pm

Pretty impressive for ISO 51200.

pgoelz · December 15, 2019, 2:05pm

Thanks to everyone who responded so far…

I am not worried about exposure or sharpening… those I have no problem with in RT and I did not manipulate them much (if at all) in the sample images.
I am concerned about the amount of work required to get an acceptable color balance quickly. If there is a way to manipulate the specific profile for my camera to get a more acceptable default color balance, I’m interested. If it requires tweaking every image individually, that would not be acceptable since it would dramatically increase the time to final image. CaptureOne seems to present the color balance (and overall appearance) of each image exactly as I see it in the camera viewfinder. RT does not.
As a related side question, why would clicking the auto levels button in RT cause most of my images (including the examples) to look grossly over exposed (with blown highlights) even with CLIP set to zero? Yes, I can then tweak the various sliders to correct the appearance, but why should I have to after using auto level? My understanding is that if CLIP is set to zero, the pixel(s) with the highest value in the image will be displayed with a value of 255 and all others <255. That does not seem to be how it works.
Yes, after some de-noising I found the high ISO image a lot better than I expected for that ISO value. I also find that I prefer the de-noise appearance in RT over CaptureOne although RT’s auto chrominance function leaves an objectionable amount of chroma noise. At least it does in that image. Interesting side fact… I also ran that image through a demo version of NeatImage (the speaker grille makes a great dark reference) and found no significant improvement over RT.

Paul

spidermonkey · December 15, 2019, 2:10pm

I haven’t looked at your file but is CA correction turned on in the raw tab? I found it changed the colour balance on high iso files in RT or ART. Just a thought…

pgoelz · December 15, 2019, 2:25pm

No, it was not checked. I just tried checking it and there was no difference.

Paul

pgoelz · December 15, 2019, 3:09pm

OK, we’re making progress. Under the RAW tab, I increased the Green 1&2 sliders in the Raw Black Points function. Under the exposure tab I switched to a standard curve and pulled the center of the curve below the midline. After some iterations I saved the profile and tried applying it to other similar images and we’re getting close.

I’ll keep experimenting from here. Thanks to all who commented!

Paul

Entropy512 · December 15, 2019, 3:32pm

I’ll have to revisit some of my old A6500 shots with RT later this week. (I’ve had an A7III for nearly two years now, so the A6500 has been a rarely used backup camera since long before I started RT.)

I know that RT has custom-generated DCP profiles for some Sonys but not the 6500 family.

Are you using automatched tone curve or the Standard Film Curve options?

If you had to change the green black point sliders than there’s a possibility that some of the black point metadata in camconst.json is wrong… It’s pretty rare that those should ever need to be touched.

ggbutcher · December 15, 2019, 5:00pm

I don’t think it’s the camera primaries, and here’s why:

A screenshot of DSC04425.ARW opened in rawproc, all that’s done here is assignment of camera primaries, subtract the camera black, apply the camera whitebalance primaries, a quick half demosaic (I use this because it absolutely does not modify any measured value), and a scale of the image to fit the display bounds:

Not yellowish like what @pgoelz is seeing. Note I’ve selected the colorspace tool; at the bottom-left, you’ll see that “assign camera profile” is selected, and the numbers used are from camconst.json, which i unashamedly have expropriated from the RT master at github. The white balance is set with the camera-supplied RGB multipliers, which are 2.617, 1.000, and 1.730.

When I open the raw in RT, I set the profile to “Neutral”, which does AFAIK the same processing, 'cept for using AMAZE to demosaic, and the image is yellowish. So, something else is going on in RT I can’t see. I can use the temp/tint sliders to get to approximate what rawproc does for ‘neutral’, and one can see the color shift in the histogram channels, so I’m inclined to think it’s a white balance thing.

A bit speculative, so FWIW…

ggbutcher · December 15, 2019, 5:06pm

Oh, btw, camconst.json says these camera primaries are from the Adobe DNG:

   { // Quality A
        "make_model": "Sony ILCE-6000",
        "dcraw_matrix": [ 5991,-1456,-455,-4764,12135,2980,-707,1425,6701 ], // adobe dcp d65

(scroll the box to the right for the comment, if your brower isn’t wide enough…)

heckflosse · December 15, 2019, 5:13pm

How about the other RT demosaickers (e.g. RCD)? also yellowish?

ggbutcher · December 15, 2019, 5:20pm

Switched to RCD, no change. Screenshot:

Now, I’m going to use the per-channel curve tools to try to align the histogram; I’ve looked, but I can’t find a place where RT exposes the multipliers… ??

ggbutcher · December 15, 2019, 5:30pm

…and with some RGB curve shenanigans:

Shown to the right is the blue channel curve; sliding the upper control point left is like increasing the blue white balance multiplier; sliding it down would correspond to a decrease. The thing I’m looking for is the histogram peaks to align; I already scooched the red curve to align it with green; scooching the blue peak to align just makes the image too blue for my taste, so I’ve left it biased to the left a bit.

I’d post the .pp3, but I think the screenshot might be more helpful…

pgoelz · December 15, 2019, 8:06pm

Not exactly sure I’m following the subtleties. The screenshot looks like the insect body is too purple… at least the way it is displayed on my laptop. Are you saying it should be correct after your adjustments? Does this accomplish the same thing that my raw adjustments did (ie., changing the greens) ?

What is the consensus so far… that RT is incorrectly decoding the ARW? And that modifying camconst.json would potentially fix it permanently?

Paul