[Bug]: Color Management seems broken

[radioproektor]

NegPy version / build

0.52

OS

Windows 10

Scanner

Digital camera (macro/copy setup)

Scanner model / details

No response

Light source

Bridgelux Thrive LED 5600K

Correction mode

Off

Input ICC profile (if any)

FFXE4 g10 IT87 F3-F130405 mtrx GS6_BpWp_D50.icc

Relevant process-tab settings

No response

What happened

It seems like there is some bug somewhere in Color Management.
My display profile is LM270WRASSA1-64F46D5F-1F80-E317-D121-DEA991498DCE.icc
My input ICC profile is FFXE4 g10 IT87 F3-F130405 mtrx GS6_BpWp_D50.icc (simple matrix well behaved profile with D50 white point, made from IT8.7 test chart as described here https://ninedegreesbelow.com/photography/well-behaved-camera-profile.html)
I just import image to NegPy and don't adjust any custom settings. Tested Raw as well as Linear TIFF.
When i enable Soft Proof, image start to look too bright, like wrong gamma was applied to it. See screenshots and uploaded examples. To compare i add screenshot of same image processed with my own method in PhotoLine image editor.

Source file

https://drive.google.com/drive/folders/1Q0leHoCdhJ1bYjXLXyc2Jd_YBeQJsvXd?usp=sharing

Screenshots

[thetalkingdrum]

The TRC of the input profile needs to be 2.2 or rather Adobe RGB (precisely encoded as 563/256, or roughly 2.1992) for this to work.

I made a branch that lets any matrix profile like your input profile (but not lut based profiles, that is more complicated) sidestep this issue, but i have not submitted it yet. @marcinz606 feel free to have a look https://github.com/thetalkingdrum/NegPy/tree/feature/icc-trc-bypass

[thetalkingdrum]

This is a quirk of how the pipeline works. I've looked into moving the OETF to the very last step so everything runs linear until the end, but that's a slightly bigger scope. For LUT-based profiles it's also kind of complicated anyway, unless they're already in the right TRC.

That one lives here but is only at a prototype stage: https://github.com/thetalkingdrum/NegPy/tree/proto/linear-boundary

The TRC branch bypass is mostly ready to go.

[marcinz606]

@thetalkingdrum looks like you already did most of the work and I just started looking :D

[marcinz606]

I'll guess I'll just unassign myself as it seems your trc branch might just handle that one

[thetalkingdrum]

Yeah, I made it, but then it got put on the back burner as the linear boundary feature was kind of overwhelming in scope (at least to me) ;)

I'll do some tests and open a PR on trc bypass when everything is ready.

[thetalkingdrum]

Seems to be working now. Fix coming in next release I guess.