Merge two categories into data driven category

[ben]

Good evening,

i am trying to create a data driven category and could need some help.

I have two categories with person names, that i would like to combine.
Both categories contain the same set of names. One is created manually, the other one by face recognition.

I am doing this, because i don't want the face recognition to write the same categories like i do manually (this was a recommendation of Mario).
But i want to see all images containing a person, independently if recognized by iMatch or manually.

Does this make sense?
iMatch doesn't provide a functionality to sync or merge the iMatch recognized persons with my manually created ones (yet).

My structure looks like follows:
_persons2
  name1
  name2
  name3
 
_persons3
  name1
  name2
  name3

My data driven category should look like:
_persons_All
  name1
  name2
  name3
 

My only solution so far is the following.
But it also creates unwanted sub-levels with other keywords (not persons):

  tag: XMP::Lightroom\hierarchicalSubject\HierarchicalSubject\0
  category filter: @All|Keywords|_Persons2;@All|Keywords|_Persons3
  use part of value: yes
  start/length: 10,100
 
Thanks
Ben

[Mario]

Wouldn't it be better to use face annotations for all persons?
You can assign a person to a face even if IMatch does not recognize the person.
And you can add a manual face annotation (and assign a person) if IMatch cannot detect a face.

Managing the same data in separate category hierarchies (some persons you manage manually, some persons managed by IMatch) will be a massive effort, and will inevitably fail. Or become out-of-sync.

And the images containing persons which you don't annotate with face annotations will never be part of person-related features, from the people view to file window attributes to sorting, searching, ...

I would give this a re-think. In the end, having a face annotation for each person in an image will be much more comfortable.

[ben]

Mario,

i totally agree. I only want to have one category of persons.
But i don't understand what would be the other approach.

Until now, i added persons via the keyword panel, which is very convenient. It highlights recently used names, offers suggestions and basically for me a fast way of doing it.
So, how would i manually create face annotation in a same quick way?
There are lots of pictures in which iMatch cannot (and neither will in the future) detect and recognise the right persons. People from the back/side/too small or whatever reason for. So, manually adding pictures to persons will always be a step in the workflow, am i right?

Thanks for your comments
Ben

[Mario]

You can of course maintain a separate person keyword hierarchy and not use the person features provided by IMatch.
I'm just not sure what if this will save you time in the end (over manually adding face annotations, which not only gives you a keyword, but also an age, integration into the People View, the People Filter, name and age display in the File Window etc.).

If you add your own person keywords or IMatch adds them when it assigns a person, there is no difference between the keywords (assuming you use the same keyword hierarchy for your manual keywords and the keywords you let IMatch assign when it assigns a person to a face).

I am doing this, because i don't want the face recognition to write the same categories like i do manually

You mix keywords and categories? Or do you use keywords? Or categories? Or do you refer to keywords as categories because of @Keywords?
Either way, a person is a person. Do you really need to differentiate between keywords/categories assigned to a file by the people feature and manually assigned by you?
Why? This complicates things a lot and produces problems you then try to solve somehow. Like in your original question above.

Maybe another user has an idea which helps to you maintain your two deliberately separate keyword hierarchies for people.
I don't recommend it.

[JohnZeman]

Because I do not have many people in my photos I don't use face recognition so I have to pass on anything to do with that feature.

But if you just want to combine the images of 2 categories why not just create a simple formula category to do that?

[sinus]

Because I do not have many people in my photos I don't use face recognition so I have to pass on anything to do with that feature.

But if you just want to combine the images of 2 categories why not just create a simple formula category to do that?

I thought exactly the same, like John.
Create simply one or several category-formulas and you are done.

I add since a long time persons with only keywords.
Keywords creates a keyword-category (what is in fact a data-driven-category - if I am not totall wrong).

Now you can create formulas without end. I do not know, what your "persons2" and "persons3" main-category means.
But if you have a clever keyword-system, then maybe you have also already in the keyword some differences between e.g. family-persons or business-persons.

For your family-members you could e.g. create favorites, this means, one click on John, another on Marie and two persons are added. Done.
I created three more favorites (keywords): alone, two and three. Maybe this can be omitted or solved better, but I do this since a long time and it works very good. Because if I want some pics from Marie, maybe I want not see Marie with a lot other people, but alone.

And now with category - formulas you can combine every thinkable combination:
John alone
Marie and John
All people
Only all family-members (specialy easy if you have a clever keyword-system like "family..." and "well-known...")
...

And I personally think, that I am very quick with this system, depending on the pictures I am for sure quicker then the AI, but in another case there is maybe the AI quicker.
Therefore I personally do not use (until now) the poeple-stuff, although very exciting.

If I would start with writing the names of people or I have only a few, I would try for sure the AI of IMatch.
It seems to be very clever and it is the future, I am sure, Mario will add in the future this and that to this cool stuff.

[ben]

Sorry, i was very busy for two weeks.
My initial question is still important to me, so i try to pick it up again.


My situation:

• Currently, i have two different keyword hierarchies (persons1|<names> and persons2|<names>). Hierarchy "persons1" is created by face recognition and "persons2" is created manually via the keyword panel
• I would like to use one hierarchy only, that includes all persons manually assigned and also by face recognition -> This is my goal and seems also strongly recommended by Mario
• For manually "tagging" faces i would like a very quick and convient function, like "recommendations, recently used, top 10". This is all offered by the iMatch keyword panel.
• Mario recommended to "add manual face annotation" for faces not detected or recognised

My questions:

• How do i add a manual face annotation? Do i have to add the face rectangle in the viewer? If yes, then that's too time consuming
• How do I overcome the risk, that "the face recognition feature" removes a keyword, that i previously assigned manually?

Thanks for your comments
Ben

[Mario]

1. You add manual face annotations in the Viewer.
it's fast, just one click. Or you use copy/paste, which also automatically assigns the correct person.

2. Assigning a person to a face adds all keywords you have added to that person.
Removing a person (or changing to another person) removes the keywords of that person, unless the same keywords also belong to another person in that image (e.g., "Family Miller").

Mixing face recognition with manually maintained separate keyword hierarchies is not an efficient workflow.
Especially not if you mix and merge the same keywords for persons and your manually maintained hierarchy.

Maybe just stick to manually keyword all your files and not use any of the person-related features at all. It's totally optional.

[ben]

Hello Mario,
thanks a lot for dealing with my questions.

Maybe just stick to manually keyword all your files and not use any of the person-related features at all. It's totally optional.

Absolutely not. I am a big fan of iMatch face recognition and was waiting for it very eagerly. I really want to use it.
I think, the great face recognition technology in iMatch had two lacks from the beginning. One was the missing possibility for batch processing faces. This has been resolved in  a great way by the face manager. The second one is a workflow for mixing manual and automatic face assignments.
There will always be persons that cannot be detected, but should be linked to a certain person (from the back, from the side, too small, too dark, ...).

Mixing face recognition with manually maintained separate keyword hierarchies is not an efficient workflow.

Why not?
If it's not an efficient workflow, then what would be a good workflow to handle manual and automatic face assignments?

Especially not if you mix and merge the same keywords for persons and your manually maintained hierarchy.

That's why i tried to combine two keyword hierarchies into one, with my initial question in this thread. But i agree, it doesn't seem a good solution.

1. You add manual face annotations in the Viewer.
it's fast, just one click. Or you use copy/paste, which also automatically assigns the correct person.

Ok, then i know what you mean. But both options are way less efficient, than what i can do with the keyword panel.
I need to go through all images in the viewer and either click and position or paste and position a manual face annotation for one person. Then i repeat the same steps for another person. Very time consuming.
With manual keywording i select several pictures in the file window and can use the many ways of the keyword panel to choose from a persons keyword (recent, top 10, suggestions, ..)

2. Assigning a person to a face adds all keywords you have added to that person.
Removing a person (or changing to another person) removes the keywords of that person, unless the same keywords also belong to another person in that image (e.g., "Family Miller").

When you mix this with manual keyword adding then you run into the following problem:
   Assume, iMatch has recognised John falsely in a picture.
   I already manually assigned "John" to the right person in this picture.
   If i unassign "John" in the face manager from the wrong person, iMatch will remove the keyword from the picture.
   I end up with a picture without the keyword "John", even though i already (correctly) added this keyword.



Please consider my ideas.
I know that other users might have different workflows.
But i really think, that having to mix automatic face recognition with manual face assignments will always be needed. At least if you work with images, that show several different persons.
iMatch offers so many time saving features, i think face recognition is worth to add some more to it.

Ben

[claudermilk]

To apply face annotations and any associated categories and keywords, I don't see that there is any way to avoid running them through the viewer. Keywords & categories get assigned when the person is assigned to an annotation, not the other way around. Then you need to go into the image to verify the correct person is assigned and the box is in the right place.

I'm in the midst of doing this myself. I have also long been hoping for this feature, and it's going to be very useful. I've had a few projects in recent years that could have benefited from it.

The way I'm handling it is by category. Like many users, I have a people tree in my category structure. I'm tackling one person at a time, which allows for some easy copy-paste of manual annotations. I am also seeing my anticipated acceleration as I move down the tree. The more files that get processed, the faster it gets because more & more containing groups have already been seen and handled, so they are a quick glance & move on. In fact last night I had one entire family group that I cycled as fast as the viewer could display the images.

For keyword & category assignments, I'm going to allow the person definitions handle those assignments for any files containing defined people. I am not seeing so far that any other keywords or categories are touched, as expected. While I haven't really specifically looked, but I'm also seeing that assignments appear to only be removed when all defined people are removed (e.g., all women assignments removed also removes the woman keyword, but if any are still in the image, the keyword remains assigned).

[ubacher]

ben says:

If i unassign "John" in the face manager from the wrong person, iMatch will remove the keyword from the picture.
   I end up with a picture without the keyword "John", even though i already (correctly) added this keyword.

This is an interesting problem which occurs when you have John assigned to the image (not to a face annotation) and you
remove an incorrectly assigned (to John) face annotation. Imatch, whe doing the delete, will not know if the keyword John was previously added by
the face recognition or manually to the image (unless it were to keep track of the source of the assignment).

How can we avoid/resolve this problem?
Adding face annotations to all existing images with keyworded people is, in my case at least, impractical.

[ben]

Ubacher:

Adding face annotations to all existing images with keyworded people is, in my case at least, impractical

How do you currently use face recognition in combination with your keyworded people?

[ben]

I opened a follow up thread, to not mix it with this question.
Maybe i come back to this thread, if i cannot find a better solution.

https://www.photools.com/community/index.php?topic=10298.0

Ben