AI mask (shape detection)

Software darktable
,
1

I was wondering if there can be a mask that can detect shapes (human, animal, bird) and can select within a selectable area. This will be a great add on, IMO.
Or one easier way is to provide a mask that can mask objects with contiguous lines.
Not sure if this was discussed already. Thanks

2

Does this fit your needs? Edge-aware bluring for blend masks

Manual:
https://docs.darktable.org/usermanual/development/en/darkroom/masking-and-blending/masks/refinement-controls/

1 Like
3

AI was discussed a lot but no one starting such discussions provided the code to implement it.
FOSS isn’t improved by discussions only, coding counts :wink:

2 Likes
4

This sure helps a bit. But I would like to avoid the initial step of ‘rough selection’ of the subject first. It would be wonderful if the subject can be selected by simple circle/ellipse object.
Even more helpful is if we can get a mask based on the initial step of ‘in focus’ plane. DT can figure out what is in focus and not in focus when I click on the ‘focus peaking mode’. It will be very useful if this technology can be used to mask in-focus/out-of-focus areas.

5

In/out of focus can be approximated using the details threshold, described in the first section of the chapter in the manual that I linked above. However, neither that feature, nor focus peaking is driven by fancy AI; they simply check for the presence of high-frequency components in the image, as you can see by displaying the details threshold control on a noisy image.

1 Like
6

Sorry if I was not clear. I understand that out of focus determination isn’t AI driven.
This is a different wish list.
Provide for a mask where I can select in-focus vs out-of-focus areas.
Ex. If I want to just blur the OOF areas (like phone portrait modes), it would be very easy to mask OOF areas as determined by existing algorithms in DT for determining focus peaking.

7

Let me re-read the document again. Hope I can understand and try to select the OOF areas in quick time. Any video would be helpful

8

Check out the masking demonstrated here:

2 Likes
9

I re-checked the masks (now I’m home and not on the phone), and it turned out that example is actually very bad, with bogus masks. Sorry about that.