Masks not very effective in 2.9 scheme script
This discussion is connected to the gimp-developer-list.gnome.org mailing list which is provided by the GIMP developers and not related to gimpusers.com.
This is a read-only list on gimpusers.com so this discussion thread is read-only, too.
| Masks not very effective in 2.9 scheme script | Ken Moffat | 10 Jul 23:38 |
| Masks not very effective in 2.9 scheme script | Ell via gimp-developer-list | 15 Jul 12:21 |
| Masks not very effective in 2.9 scheme script | Ken Moffat via gimp-developer-list | 15 Jul 20:08 |
Masks not very effective in 2.9 scheme script
Hi,
TL;DR - using masks in a 2.8 scheme script works as intended to extract parts of an image, but after converting it for 2.9 it does much less.
One of my interests is photos - particularly trains, and for that I like to resurrect as much detail as possible. Fortunately, I've been taking raw photos for several years. Towards 2 years ago I found out how to use three versions of the same raw image : correct exposure (for the intended subject), overexposed to add shadow detail, similarly underexposed to restore the highlights. For the overexposed image I make a mask from itself, desaturated and inverted, and for the underexposed one I just desaturate the mask.
At that time I was doing it all manually (and that still worked fine when I last tried it in 2.9.4). There is of course a problem with this approach which may make it unsuitable for some uses - the resulting image lacks contrast, but there are ways around that.
Eventually I decided that I wanted to script this. By that stage I had started to use 2.9. I got the script working ok on another machine using 2.8, but after making changes for 2.9.4 it seemed to do almost nothing. At that stage I was still hoping to make the masks from the "correct" exposure (it sounded like the right thing, but turns out to give "too much of everything"). I then simplified the 2.8 script, used it, decided it was ok. Somebody sent me fixes for 2.9 (change NORMAL to NORMAL-MODE, and gimp-desaturate-full to gimp-drawable-desaturate).
The 2.8 version, and some examples (with a faulty process! - I had assumed the G'MIC tone mapping was similar to Advanced Tone Mapping, Doh!) are at http://zarniwhoop.uk/three-exposures.html - the plugin for 2.8 is linked from the bottom of that page.
BUT (at last!) this still does much less than in 2.8. I've been using git 14795c1f72 from 26th June. By turning off the undo group I've been able to look at steps along the way. The problem shows in the attached 800x602 example pngs when I do the shadow layer: The overexposed image, and the desaturated and inverted images I'm going to use as the masks look the same, but when I apply the mask not very much happens.
Originally I was going to attach some 800x602 pngs, but the list would not let me do that (mail too big), so instead I've created a temporary page showing the good 2.8 side by side with 2.9 :
http://zarniwhoop.uk/problem.html
The files are:
x-2.{8,9}-three.png - what the plugin produces. In 2.9 it is not very different from the base image without the plugin.
x-2.{8,9}-shadow-mask.png - the masks for the shadow layer (i.e. for the overexposed image). These appear to be similar, so the different desaturate command is probably not an issue.
x-2.{8,9}-overexp.png - the overexposed image, after masking, with the other layers deleted and then superimposed on a white background to see it more clearly. In 2.9 there is seems to be less of everything.
I've also put links to the 2.8 and 2.9 scripts at the bottom left of that page.
I notice that the file sizes of pngs from 2.9 are a lot bigger, but I'm sure that you already know that.
From looking at the script you can probably tell that I don't understand scheme, so maybe I'm doing things wrong - this was just an attempt to convert my manual steps into a script.
Any suggestions, please ?
ĸen
I live in a city. I know sparrows from starlings. After that everything is a duck as far as I'm concerned. -- Monstrous Regiment
Masks not very effective in 2.9 scheme script
On Tue, 11 Jul 2017 00:38:17 +0100 Ken Moffat wrote:
Hi,
TL;DR - using masks in a 2.8 scheme script works as intended to extract parts of an image, but after converting it for 2.9 it does much less.
The most likely reason for the different results, is that in 2.9 masks use the pixels' linearized intensity as the alpha value, while in 2.8 they use the gamma-corrected intensity. The linearized intensity values are generally lower than the gamma-corrected ones, so the resulting mask values, after pasting a desaturated image into the mask, are lower in 2.9 than they would be in 2.8, resulting in a subtler effect. Note that these sorts of interactions is something we're still working on.
For now, you have several options: You can cast the mask values from gamma-corrected ones to linear ones. While there's no "obvious" way to do that, there are several hackish ways. For example, you could use "colors -> components -> extract component", select any of the RGB components, and enable "linear output". You can do that either before or after pasting the desaturated image into the mask; either way, you should make sure to use high precision ("image -> precision"), preferably floating point, or else there'll be notable data loss in the intermediary results, in either the dark or the bright areas. Unfortunately, this operation is not currently available to scripts. You can get a good approximation for it using the levels tool instead ("colors -> levels"), with a gamma of 2.2.
Alternatively, you can avoid masks and use other forms of layer compositing, although this might be more cumbersome. Instead of attaching a mask to a layer, keep both the masked layer and the mask as ordinary layers, and put them in their own layer group, with the "mask" layer below the "masked" layer. Use "colors -> color to alpha", with black color, on the mask, to turn black into transparency. Then, set the upper layer's mode to "multiply", or, alternatively, to "normal" (from the default group), while also setting its composite mode to "source atop" (in the layer attributes dialog). If you want to keep the mask layer opaque, then instead of using "color to alpha", add a black layer above it, set its layer mode to "color erase" (from the default group), and set its blend space to "RGB (perceptual)" (in the layer attributes dialog).
All of the above (except as noted) is doable through a script, although I'll leave it to you to work out the details :) The procedure browser is your friend.
--
Ell
Masks not very effective in 2.9 scheme script
On Sat, Jul 15, 2017 at 08:21:32AM -0400, Ell via gimp-developer-list wrote:
On Tue, 11 Jul 2017 00:38:17 +0100 Ken Moffat wrote:
Hi,
TL;DR - using masks in a 2.8 scheme script works as intended to extract parts of an image, but after converting it for 2.9 it does much less.
The most likely reason for the different results, is that in 2.9 masks use the pixels' linearized intensity as the alpha value, while in 2.8 they use the gamma-corrected intensity. The linearized intensity values are generally lower than the gamma-corrected ones, so the resulting mask values, after pasting a desaturated image into the mask, are lower in 2.9 than they would be in 2.8, resulting in a subtler effect. Note that these sorts of interactions is something we're still working on.
Thanks for the detailed reply. I'll have to get back on the 2.9 machine at some point to work through the options. Meanwhile, I've saved a full copy of your mail in my notes.
Cheers.
ĸen
I live in a city. I know sparrows from starlings. After that everything is a duck as far as I'm concerned. -- Monstrous Regiment
