Weird phenomenon with Panasonic G9 and Darktable

I noticed something very weird. When I take a photo of a very bright area like the sun with my Panasonic G9 on ISO 100, there’s a strange pink thing in the sun. Here’s a sample Raw:

http://jpolak.org/downloads/PANA9074.RW2

It doesn’t happen at base ISO 200, only with the extended ISO 100 value. Here is a screenshot in darktable:

This has never happened with any other ISO in the “real” ISO range.

Do you have highlight reconstruction on?

1 Like

Wow, fast reply! Thanks for the easy fix! It was on, but if I just decrease the threshhold a little, the purple stuff goes away. Nice!

I vaguely recall something about variable black levels… Not sure if its the g9 or another model. A search of the forums might reveal.more to you :slight_smile:

It’s the white level, if I’m not mistaken. But yes, in some cameras it change depending on the ISO.

@JasonTheBirder You should report it on GitHub, so it can be fixed.

Following up on @Donatzsky you might want to check what raw white value DT uses and what is reported in the exif data…that has often in the past shown where the disconnect is…

Thanks, Todd. Is that something I could do myself?

Sure. I use xnview and I use that to browse the meta data of an image… I tried to download your image but the link didn’t work… I can do a quick check if you update the post and upload the image

Just as an example you will see something like this…

Looking in DT you will see

I think the black levels don’t match because I think there were issues for some panasonic camera’s and the black level so by default its bumped a bit …but this is the sort of check I was talking about…

EDIT…I found g9 images on imaging resources…shot at 100 and 200 iso…

WP is 4095 for 200 iso but for 100 the data shows it at roughly half (2111) …but DT still uses 4095 so you could make that change and see how your images look or confirm its the same as what I found…

If I remember correctly, ISO 100 is an ‘extended’ white level on those cameras: the camera still shoots at its base ISO, but overexposes the image, then divides the result by about 2.

In this example I’ll use 2 as the divisor, and assume a 12-bit DAC, meaning a max recorded value of 4095.

Suppose your scene would require ISO 200, f/5.6, 1/1500 s or ISO 100, f/5.6, 1/800 s. The camera will shoot with 1/800 and ISO 200 (its native ISO) instead. Now, everything is twice as bright as it should be. Therefore, the camera divides the numbers by 2. This way, both the signal and the noise are reduced; the max signal value will become 2047 (the maximum with 11 bits). However, any highlight that was blown out because of the overexposure (clipped to 4095) will remain blown out (clipped at 2047), division won’t help with that.

Now, if darktable used 2047 as the white level, it’d scale the values back up, bringing the brightness up again. So, I think what happens is it needs to use 2047 as the white point for highlight reconstruction only, but then pretend the white point is 4095 to keep the recorded values where they’d be on a camera with native ISO 100.

I don’t know if that makes sense and if it’s correct.

What happens with the black levels is then a question of that in-camera magic, I guess.

(edit: fixed minor typos, I had typed this on the phone)

2 Likes