Selection Behaviour

Software GIMP
1

Hello,

I’m using Gimp 2.9.7~25-ubu16.04.3~ppa~5a28753 from otto-kesselgulasch ppa on Linux Mint 18, because I need the 16bit image feature.

I was trying to do the Luminosity Masks tutorial from @patdavid. I used the same image as in the tutorial.

But when I create the first Dark channel I got very different result than the tutorial. The mask is lighter than expected.

Dark_channel
Dark_channel1920×1440 256 KB

In the tutorial, it says that the first Dark Mask should be the invert of the Light one. If I try to invert the Light one I get a totally different looking mask (see below)

Light_channel
Light_channel1920×1440 615 KB

I tried to follow the tutorial on a previous version of Gimp (don’t know which one) and it works as expected.

It’s look like to me that the subtract selection feature behaviour have changed in the newest version

Does anyone experience the same kind of trouble ?

Thank you for your help

1 Like
2

@Thomas Welcome to the forum! I had questions about the tutorial a while back. See: Generating mid-tone masks. If you have any more questions, I would be happy to help.

3

Thank you for the link. It helped me better understand the inner working of the Mask and like you why my MM and MMM mask were always completely black

But I still think I have an issue with the way selection difference is handled in my version of Gimp.

I took the file with the gradient and “redid” Light and Dark mask. The difference between the original and mine is pretty obvious but I have no clue why.
lum-test-thomas.xcf (353.6 KB)

4

I suspect something is going on, but I’m in the middle of another thing at the moment and can’t test it. :frowning: I’ll test against my 2.9.6(ish) and report back. I’m holding off on updating the tutorial there until we push a 2.10 and things have settled down a bit. :slight_smile:

5

I will let @patdavid assist you since he is da man!

6

There is a solution to that inverting issue. I believe you can use G’MIC, and somehow flip the L in LAB to get the dark mask. I don’t use GIMP though, so I can’t really say that’s the temporary workaround, but flipping the lightness would probably be the key to the answer.

7

I did this with Gimp (invert mask) as a workaround but if I try to masks LL or DD as describe in the tutorial, result start to “drift” from the expected.

8

I tried GIMP for a while to see if I can get around it, but I disabled image color managed display, and it seem to work as expected.

9

Which version of GIMP are you using ?

10

At the moment I’m still looking into this with the devs. In 2.8 the selection works as expected, but not so much with 2.9.6+. I think the problem is due to the operation now being performed in linear light as opposed to gamma corrected.

I have been testing with this XCF: gradient.xcf (386.2 KB)

In 2.8 it selects the mid-point of the gradient, while in 2.9.6 it selects about the brightest 25% of the gradient:

1000×257 156 KB

This was what I was told by tmanni in #GIMP:

patdavid_m: if I’m not wrong, “medium gray” value in floating point precision is 0.5 in gamma corrected and something like 0.215 in linear light

the issue is here app/core/gimpboundary.h · 70881728f2495ac82d606d676602861ffb176b50 · GNOME / GIMP · GitLab

2 Likes
11

Thank you @patdavid for following on this issue. :+1:

I didn’t expect that you will move forward and report the bug to the gimp dev team. I was ready to do it. My question was to make sure that was a bug and not feature before doing so :grinning:

Let me know if I can do something to help on this matter.

1 Like
12

It was pretty easy since I’m around them every day anyway. :slight_smile:

My bug report is here:

https://bugzilla.gnome.org/show_bug.cgi?id=791519

And tmanni also opened something for the UI/UX stuff as well (the marching ants of the selection):

https://bugzilla.gnome.org/show_bug.cgi?id=791512

And more importantly, thank you @Thomas for catching this and bringing it to our attention!