Enhanced parametric masks

[brucewilliamsphotography]

Hi devs, Here's a situation I run into a lot. I can't construct a complicated mask using parametric mode, because the area I wish to select contains too wide a range of luminosities, too wide a range of hues, and too wide a range of saturation values. But often, if I was to break the selection down into smaller chunks, I COULD make valid selections using smaller ranges of Jz, Cz and hz, (or L, A, B, C, h) in combination with a drawn element. So I had this idea... what if you could define a drawn+parametric mask, and then "dump" it to something like an alpha channel mask (to use potatoshop terminology) which could sit in the Mask Manager module alongside any drawn masks. Once this D+P mask had been added to the Mask Manager as an alpha mask, the parametric values would be cleared, allowing the user to define ANOTHER parametric mask using different parameter values. In the perfect scenario, you could do this multiple times, defining small portions of your image with the parametric mask sliders, turning them into alpha channel (bitmaps?) masks, until you had assembled all of the masks required to construct your one big complicated mask, which could then be achieved using the combining modes in Mask Manager. Am I making sense? Does this sound achievable? Would it be of any value to anyone else?

[ralfbrown]

Probably the most straightforward way to implement that would be to allow a raster mask as a "shape" on a drawn mask, i.e. in addition to being able to select "use same shapes as" one would be able to select "use same mask as". With that setup, you would normally want the earlier module that generates the raster mask to be a no-op, e.g. an instance of the exposure module set to 0 EV.

The way drawn masks work is that an individual mask is generated for each shape, and those are combined with the specified operators (AND/OR/sum/etc) to generate the full drawn mask. So a raster mask would simply be another individual mask that happens to have been created by another module instance rather than from a user-specified shape.

[todd-prior]

@brucewilliamsphotography I'd be curious to see the image and the regions that you speak of just to fully understand the task at hand... usually you can put these together with polarities etc and manage to mask things...

[brucewilliamsphotography]

Todd, I'm at work at the moment, but will try and dig out the image that inspired this request when I get home tonight. I'll chuck it online somewhere and post a link.

[todd-prior]

Prior,Todd reacted to your message:

From: Bruce Williams @.> Sent: Monday, January 29, 2024 12:28:33 AM To: darktable-org/darktable @.> Cc: Prior,Todd @.>; Comment @.> Subject: Re: [darktable-org/darktable] Enhanced parametric masks (Issue #16188)  

Caution: External email.

Todd, I'm at work at the moment, but will try and dig out the image that inspired this request when I get home tonight. I'll chuck it online somewhere and post a link. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.***>

[brucewilliamsphotography]

This was the image I was trying (unsuccessfully) to mask.... http://www.audio2u.com/download/darktable/_DSC4580.ARW I wanted to isolate Teigan from everything else.

[TurboGit]

It really depends on what you want to do with Teigan, if only adding for light or some "light" edit a simple curve could do:

And with a curve to add some exposure:

Before:

After:

_DSC4580.ARW.xmp.txt

EDIT: Was just a quick test done in less than a minute.

[brucewilliamsphotography]

So you're just using a drawn mask only, Pascal?

[TurboGit]

Yes, two actually. One to get the whole person and one just to remove the empty part between the hat/arm/shoulder/head. Then a little use of the guided filter and that's all. In photography it is very rare to have a need of a very precise (at the pixel level) mask.

[TurboGit]

Let me digress a bit :)

That's why I do not see a need to IA based selection for photography. It may be needed to composing if you want to cut&paste objects from one image to another in GIMP or some similar software (no name :). But in photography we just need to have smooth selection to avoid seeing the change at the border, and controlling this smoothness part is what guided filter is for. And all my masks have been done in less than a minute since I'm using darktable, so I'm not sure to see what an IA based selection could bring here. I'm not even convinced that in your example the IA would properly select Teigan.

[todd-prior]

I'm not even convinced that in your example the IA would properly select Teigan.

I'll check when I go home how good ON1 is. I have that product. I have never stuck with it because I find the tools in DT more to my liking and I spent a lot of time learning them but recently I did run a set of images through it and AI masking is not bad actually is helpful... You can make a preset for say foliage and it will mask what it thinks is foliage and it generally does a good job and the power comes from the concept of dymanic masking ie now you paste your edit on a photo and it will adjust the mask to what foliage is there or what sky is there etc and so it accounts for changes from image to image.....

EDIT:

This was the mask without any tweak of the sliders on the tools..it missed a bit of the hat... you can easily refine the edges but I just offered it to show what a proposed AI product would do by saying select people..

This is the image opened in ON1 and then a BW filter applied to show the unedited mask....

[brucewilliamsphotography]

Pascal, I wasn't suggesting any kind of AI tool in my initial request. I was suggesting (crudely!) the desire for some way to build a mask up bit by bit. But I understand your point about drawn masks. They can be fairly quick to implement. I just sometimes find the control points on drawn masks VERY fiddly to grab with the mouse, which makes varying the feather around different parts of the masked region quite tiresome to implement. :)

[todd-prior]

@brucewilliamsphotography Try grabbing segments of the mask between nodes ..they are easy to grab and bold up nicely. I find dragging them a little in or out and having a decent feather can allow you to easily add a bit more or take away from a mask... the opacity contrast can really help too... if you keep in from the edge you can get it to bleed out nicely

[brucewilliamsphotography]

Todd, that's the bit I have trouble with, though! I find the feather nodes, as well as the mask nodes themselves, incredibly tricky to capture with the mouse. That was the case when I only had a 1080p monitor, and is no different now with my 4k monitor. Guess I just have to keep practicing! :)

[todd-prior]

Todd, that's the bit I have trouble with, though! I find the feather nodes, as well as the mask nodes themselves, incredibly tricky to capture with the mouse

I feel like I changed the color by picking the gridline color I liked the most and I may have found a way to make them thicker in css or other but maybe not ...I know it did do something a ways back...

Anyway if you hover should be pretty easy...

https://github.com/darktable-org/darktable/assets/44409419/d3e37c01-245f-4125-8c5e-efd91e1ef5f8

[github-actions[bot]]

This issue has been marked as stale due to inactivity for the last 60 days. It will be automatically closed in 300 days if no update occurs. Please check if the master branch has fixed it and report again or close the issue.