Artifacts in High-ISO CR3 from EOS R7

Hi there,

trying to develop a series of high-ISO shots taken with my EOS R7 in cRAW, I recognized some strange artifacts/areas in the developments. See attached screenshots that should show this clearly.

Interesting thing is, that developing a DNG file generated from the very same Raw via Adobe DNGConverter does not show this behavior.

Has anybody here experienced such a thing before? I would assume, that this is an issue of the used raw-development engine …

Lightroom Classic behaves nicely as desired without any issues with the Raws…

Any ideas?

Best regards, Martin


cRAW import is handled by the LibRaw engine. You could try testing just the LibRaw decoding to rule it out first, perhaps via the unprocessed_raw test program they ship (you’ll need LibRaw 0.21 installed on your system).

Hi,

not sure if I do it correctly…
But attached what was the output from unprocessed_raw.

As I see, no obvious artefacts/strange regions.

What are the artifacts…you added an exposure bump to the second image…what do they look like at the same exposure and short of the noise and maybe some issues above the number on the singlet I didn’t notice something clearly different but not sure what I am looking for??

That’s a good data point, thanks.

What about if you go all the way back on history stack to “0 original”?

I assume the weird blurred regions. Is that right, @kmilos ? → sorry, I meant to ask @DerMartin.

Right, high frequencies missing, but weirdly only in some blobs…

1 Like

If the white point is wrong, highlight reconstruction may think those areas are overexposed and may try to paint something in? What happens if you disable that module for now?
Another thing you could try is the demosaic module (use a different algo).

cRaw is not Bayer AFAIK, so it shouldn’t have an effect?

I can change demosaicing method on craw from my R6. VNG4 for deep shadows due to compression artifacts.

Any raw files to share, @DerMartin?

If it’s not mosaicked data, why is the module active?

I’d guess the libraw integration isn’t a robust as rawspeed.

Either way: can the demosaic module be disabled, or if not, switched to another algorithm?

Can highlight reconstruction be disabled?

Certainly, but likely requires a bug report

Hi,

sorry, was away from my PC for some hours.

Yes, I am talking about the weird blurred regions.
I would also assume this is related more to libRaw than to darktable-internal processing.

If I go back to “history point 0” the regions also are visible.
Here two 100% screenshots of another image in which I see this happening:
“History step 0”

Developed Version:

Attached also the respective raw file:

20230121-173420_R7MA6341.CR3 (26.5 MB)

This file is licensed Creative Commons, By-Attribution, Share-Alike.

About demoisacing. It acts like it should.

https://raw.pixls.us/getfile.php/5634/nice/Canon%20-%20Canon%20EOS%20R7%20-%203:2.CR3

No issue with your file for me.

Interesting. What is your setup? dt-Version? OS? Graphic-Card?
I have 4.2.0+106~gdba8f7bda1 (build from git), on manjaro-Linux with NVidia driver.

Strange indeed …

This is my old laptop from 2011 that almost doesn’t work.
Ubuntu 22.04. No OpenCL.

I installed 4.2 for 20 minutes ago from here Install package graphics:darktable / darktable

XMP file if you want to try my settings.
20230121-173420_R7MA6341.CR3.xmp (4.8 KB)

Odd. It’s fine for me too:


Win 11, dt 4.2 from darktable.org, openCL enabled on GTX1650