Face Recognition - People AI / Option for "not relavant anymore"

[vsammy]

Hello Mario,

while tagging more and more photos with faces on it, I wondered if you could add an option to exclude people who aren't on newer photos anymore. Something like the death-date you have already added, where your AI don't suggest these people.

There are many people on old pictures who aren't "relavant" anymore for newer photos, becaue they e.g. moved or whatever. Wouldn't be that "easier" and more acurate for the AI if there was an option/flag in the people manger for that?

Regards
vsammy

[Mario]

IMatch considers the lifetime of persons when finding persons for faces.
Do you have a case where IMatch assigns a face to a person which was already deceased when the image was taken (assuming the day of birth and day of death and image date are correct)?

Adding a "Don't consider for automatic assignments anymore" would be possible. But I wonder how much good would that do.
It would only help if you often encounter situations where you have persons with similar faces, but only one of them should considered anymore (and is not dead)...?

[vsammy]

Hi,

[quote
Do you have a case where IMatch assigns a face to a person which was already deceased when the image was taken (assuming the day of birth and day of death and image date are correct)?

[/quote]

It seems that it works. And that's why the idea to add an flag or something like ,,Don't match any newer faces to that person"...

[quote

Adding a "Don't consider for automatic assignments anymore" would be possible. But I wonder how much good would that do.
It would only help if you often encounter situations where you have persons with similar faces, but only one of them should considered anymore (and is not dead)...?
[/quote]

Yes, at least it could help if you have some 150 Persons and about 50 are not in you ,,life" anymore....

Regards
Vsammy

[Mario]

You have 150 persons and 50 of them are not relevant but look very similar to the other 100 people?
I assume that this can happen, but for how many users? Let me know below.

Please, when you quote, quote correctly. This makes your posts more readable.

[vsammy]

Sorry for misquoting, I tried....

I thought this forum area is for ideas that you might be able to implement, so here again my wish without claim to implementation.

To describe again my idea: Yes, I have about 150 faces of about 320 people, of which about 40-50 irrelevant faces are assigned to the same 150 people over and over again. Apparently the AI sees some similarity.

Of course I don't know if it is in the interest of the other users. In my case I have persons from several decades in my database.

Pictures from studies - these people are not here anymore for a long time and there are no new photos - but to mark them as "died" is unclean(would go if necessary though...)

Children's pictures from school, kindergarten, acquaintances from back then..... all still alive, but there are no new pictures of them today.

From this came the idea to implement a similar marker as "died on:", so that the AI does not assign the faces to these people anymore. This works very well, as I think with the "died on:".

As I said: just an idea. Because maybe it makes sense for some.

[Mario]

This forum is for asking for new features and changes.
I usually don't comment, unless I see technical problems, the feature already exists or there is a way to achieve the same result.

Since only a part of the IMatch user base participates in this community, every +1 counts.
If a FR gets a number of +1 from other users fairly quickly, I look into it.
If not, the feature is of no interest to others.

I would recommend you check in the Face Manager if the faces of the unwanted persons are in the trained set for a person. In that case, un-training the face in the Face Manager should work just well.

To implement your feature I would need to enhance the database model, IMatch Anywhere WebServices, the Person Editor, documentation, translations, tests etc.
But, most important, I would need to add additional checks inside the tight and optimized to the hilt routines which perform face recognition.
Making even small changes on this level can cause substantial performance degradation, for which I then need to figure out a solution.
Problems like this are not uncommon - but way too much work and cost for something only one or a handful of users will ever need.

[vsammy]

Thanks for explaining, as a non-programmer I wasn't aware that it was such a work to program that. So I understand that if this is not implemented. As I said: should only be an idea that might have affected several users.

[vsammy]

Hello Mario,

the Workaround with "un-training" the faces seems to work. The AI seems to "ignore" theese persons afterwards.

Regards,
vsammy

[Mario]

That what I had assumed. A untrained face is considered less relevant than a very similar but trained face.