Face recognition combined with manual person tree

Started by ben, May 23, 2020, 10:17:53 PM

Previous topic - Next topic

ben

Hello everyone,

this is a follow up to another question i had using data driven categories.

I hope to get some better advice and more ideas about how others are solving the following problem.
I am close to giving up, but don't want to, yet.


Question:
How do you handle your manually created "persons-tree" (using keywords or categories) in combination with the new face recognition and the "people" view?


The challenge:
I already asked several questions, that finally lead to the above one. I now think, that's the core of where i am stuck.
I love the new face recognition and absolutely would like to use it. But currently i have:

  • faces recognized by iMatch in the people view
  • faces not recognized by iMatch (from the back, too dark, too small, ...)
  • persons from a 30 year time range, which i have categorized in a keyword tree (persons|<name>)


My ideas / what i need:

  • I only need the information that a certain person is in the picture. Having the information "where in the image the person ist located" is nice to have, but not needed
  • Adding the information that a certain person is in the picture should be really quick and should work with many images selected. This working step should offer easy access to person-tags/names, e.g. recent-, top 10-, suggestion-, favourite-names
  • Using a manual face annotation would mean i have to go through all pictures and copy/paste or add a face annotation and position the rectangle. This is way to time consuming in my eyes



Thanks for taking your time.

Ben







Jingo

I think it depends upon your ultimate goal for using face recognition and the results.... are you interested in automating the keyword process or are you interested in securing face annotations (or both)? 

I've thought long and hard about this since first testing out the functionality... for me, I already have thousands of photos with keywords for family and friends... most of the images span 10-15 years of digital photography.  I have no need to use face annotations because I know the 2-3 people in each image and the keywords are clear and easy to detect.  I also am not adding in additional photos from years and years ago... so, for me, it would only be used to reduce keyword adding to future photos.

In my case, I trained the system on new images for family and friends.. and as I import new images into the DB, the faces are recognized and keywords added appropriately.  It saves me a bit of time so I use the function.. but this will only be done going forward since I am not really interested in the annotations.  I can also change my mind and run recognition across an entire group of images for a person if the future provides me some reason to use the People feature in more detail.

Hope this helps...

sinus

Quote from: Jingo on May 23, 2020, 11:52:20 PM

... for me, I already have thousands of photos with keywords for family and friends... most of the images span 10-15 years of digital photography.  I have no need to use face annotations because I know the 2-3 people in each image and the keywords are clear and easy to detect. 

Here I am in the exact same boat.

And also for business I do not need something from this people-stuff (though fascinating).
And to be honest, I GUESS, with a clever workflow, e.g. created some Metadata-templates and add these then to favourites, I am not slower then use the new AI, I would even think, quicker.

But I think, this feature from IMatch will even be better and better and will have this and that new feature, hence I can imagine, that I will use it in the future.
If I would start a new DB, I think, I would use it up from the beginning.
Best wishes from Switzerland! :-)
Markus

ben

Quote from: Jingo on May 23, 2020, 11:52:20 PM
I think it depends upon your ultimate goal for using face recognition and the results.... are you interested in automating the keyword process or are you interested in securing face annotations (or both)?  [..]
I have no need to use face annotations because I know the 2-3 people in each image and the keywords are clear and easy to detect.  [..]
In my case, I trained the system on new images for family and friends.. and as I import new images into the DB, the faces are recognised and keywords added appropriately.  [..]

Thanks, i think we are looking at it the same way.
I mostly don't need the face annotations themselves, only the information that a certain person is in the picture (hope i understood you right). For some family pictures it's nice, though, to see the face annotations directly linking the names to a face.


Quote from: sinus on May 24, 2020, 09:43:08 AM
But I think, this feature from IMatch will even be better and better and will have this and that new feature, hence I can imagine, that I will use it in the future.
I agree, i don't want to abandon the face recognition and also not the people view, because i think it will be more and more important in the future.


Question 1:
What do you think of the following proposal?
The people view could show all faces with face annotations and also those from a, user definable, category- or keyword tree.

Question 2:
In case you think, the proposal is making it messed up:
Do you think, we could need another link between people with face-annotation and people from manual category- or keyword-tree?

Ben

Jingo

Quote from: ben on May 24, 2020, 11:47:04 AM

Question 1:
What do you think of the following proposal?
The people view could show all faces with face annotations and also those from a, user definable, category- or keyword tree.

Question 2:
In case you think, the proposal is making it messed up:
Do you think, we could need another link between people with face-annotation and people from manual category- or keyword-tree?

Ben

I had thought long and hard about this while testing out the feature early on... but since annotations are at the heart of the People feature (and keyword linking is just a nice feature) - I don't think the People feature can function well without them.  My idea was to write an app that just added a "dummy" annotation box to all images for a person keyword (creating a Person entry if needed even)... but Mario pointed out this would not be a good idea because it could break the system and cause issues down the road... so I decided against it (also, there are no annotation endpoints available so I'm not sure I could actually do it without getting creative).

In the end - I decide to just take a wait and see approach! Like Markus, I can just as quickly add keywords to my images for family members using metadata templates, favorites, and the keyword panel.... so the AI is great and helpful.. but not really for my workflow.

ben

Ok, so basically you keep going with your own keywords.
And face recognition is just a nice ways to add keywords.

Jingo

Quote from: ben on May 25, 2020, 06:05:42 PM
Ok, so basically you keep going with your own keywords.
And face recognition is just a nice ways to add keywords.

Yes.. that is correct...for me anyway..  :-)