How to convert Crosstalk Matrices slider values for normal Channel Mixer range -100% 100%?

[radioproektor]

Currently NegPy Crosstalk Matrices slider use values range from - 0.5 to 0.5. This is rather unusual because in most other apps used range in % values like -300% 300% in PhotoLine or -200% 200% in Photoshop.
So i just want to ask - Is it any simple way to convert NegPy Crosstalk Matrices slider values to Channel Mixer percent values?

P.S. I am rather skeptical about this crosstalk compensation because in PhotoLine instead Channel Mixer i simply move C/M/Y hue values Hue Editor and this fixes strange looking colors. Also to be honest i am still not sure if this problem really due crosstalk or due invert in non uniform color space which produce hue shifts by its nature.

[radioproektor]

Seems i find answer in manual https://github.com/marcinz606/NegPy/blob/main/docs/CROSSTALK.md
NegPy value 1 is equal to 100% in other apps Channel Mixer.

So generic C41 matrix will look like this:

    R      G      B
R  100    -5     -2
G -4       100   -8
B -1      -10     100

I applied it like this (but for me correction was right only when used positive values) and got results very similar to those i manually correct with Hue Editor.

    R      G      B
R  100     5      2
G  4       100    8
B  1       10     100

Yellow shifted to Warmer red, Blue shifted to Cyan, Less magenta in Reds. It also produced less Cyan in Greens (but to be honest i personally prefer to keep greens untouched or even slightly increased).

[radioproektor]

I just checked how Crosstalk correction works in NegPy, and i have a subjective feeling that it works in opposite direction. Currently it increases incorrect color shifts instead of suppress them. Please check if it is a right idea to change Crosstalk Matrice negative values to positive and positive values to negative. I can create a new issue ticket if it helps.

[radioproektor]

Here is test of my theory in NegPy.
Correction disabled

Default C41 correction (subjectively looks bad)

Custom C41 correction with negative values changed to positive values (looks ok)

[thetalkingdrum]

Hi, just wanted to clear up the first question: The crosstalk matrix works in density space before the image is inverted. It is not the same as the channel mixer in photoshop or photoline, and uses completely different math, so there will not be any slider value you can copy and it will give the same result. Even your 1.0 = 100% relabeling is just renaming the number, not converting the effect. The same value does something different in density space than it does in a channel mixer.

[radioproektor]

Well, for me it seems like works with Channel Mixer In Linear gamma as well. I also tested it in density space (linear-to-log->channel mixer->log-to-linear) and it also works. I didn't try to adjust values manually, but i guess it make sense to fine tune it a little bit.

[thetalkingdrum]

Yeah, tuning is really the whole point of the crosstalk mixer.. There's no single "correct" matrix, generic or made from some film stock, it's meant to be dialed in for your own film, light and sensor. If your fine tuned version looks right on your images, it is working as intended. Then you can save that as a matrix and continue using that matrix on your rig.

[thetalkingdrum]

on the second part: there is no bug in the code, I checked. The crosstalk matrix isn't a dial with one correct direction that always fixes color shifts, it is a per rig calibration. Generic C41 is just a datasheet approximation for the dyes only, your light source and sensor add their own crosstalk it doesn't know about, so the values that fit your setup can look different, even flipped. That's not a bug, it just means you need to tune your own matrix instead of using the built in one.

[radioproektor]

Well, i tested few other examples from other setups shared online and they all have that yellow color with too green tint. And default crosstalk correction in NegPy always adds more green tint to yellow colors. Same time crosstalk with positive values makes yellow color look like normal yellow.
For example you can test yellow bus photo fropm NLP sample RAW files https://negativelabpro.s3-accelerate.amazonaws.com/sample-raw-scans.zip

[thetalkingdrum]

Default crosstalk correction in negpy just adds some color separation mostly. It says nothing about your rig. You need to tune it so the yellow looks good. Then you have a matrix that works for the set up that captured the bus. But not your own set up.

[radioproektor]

[radioproektor]

I noticed one more possible global confusion in Channel Mixers. Input/output channels could be located in rows or columns in different apps and datasheets. For example Photoshop vs NegPy table placement.