Most efficient way to clean up false positives in face manager

[dkyndt]

Hi there,

I wonder what the most efficient way is to clean up false positives in the face manager.
for example, I have 2500 photo's where one person in the photo is person X.
I'm pretty sure that not all those photos have person X in it.
Some of the persons have been allocated to person X by the software, some of them are human error, where I clicked the wrong person.

Is there a variable "percent equal to others" available in Imatch? Imatch proposes other persons like person A, person B, Person Z.
If this should be available, I think it would make it easy to sort out the variable al check the photo's with the lowest equal value.

BR Davy

[Mario]

I'm not sure that I follow. An example would be nice...

The AI produces person matches with a 'similarity' value. This value is for example used by the Person Selector to list the best possible matches for you to choose from.

The Face Manager works with one specific person (X, in your example).
It shows all files with faces where the AI thinks that the file contains the person in the unconfirmed view.
This means that the "best match" for the given faces is person X. This is why IMatch has assign X to the face and not Y.

What would persons equal to others do?

IMatch uses a sophisticated clustering algorithm for the People Organizer to build groups of similar faces.

[dkyndt]

Hi Mario,

I know the software uses a neural network to calculate the strength between a face and a known name. (the similarity value)
Sometimes it misses, or I also made some mistakes in assigning a face to a name.

Lets say we have 5 photo's with 2 different faces, and 3 names A to B.

Photo 1 : FaceA ==> Name A
Photo 2 : FaceA ==> Name A
Photo 3 : FaceA ==> Name A
Photo 4 : FaceB ==> Name B
Photo 5 : FaceB ==> Name A

I have 3 faces, corectly assigned to name A.
I have 1 face correctly assigned to name B.
I made one mistake on face 1 and assigned it to name A.

you can rectify this, although I have several names with over 500+ photos attacted to it.
Therefor it would be a very intesive job to do this manually by checking all the photo's and see if the face matches the photo.
These error makes it harder for the program to propose a correct name to the face.

Is there a way to cleanup these mistakes? Is the "similarity" variable available to filter or sort?

[Mario]

When you replace a person assignment (to fix your error), IMatch automatically re-evaluates all non-confirmed and unassigned faces in the database.
This usually fixes such issues automatically. The AI forgets the old person assignment and assigns the now best matching person.

You may also want to check in the Face Manager which faces have been trainned for the person and remove unwanted ones. Maybe train some additional faces to strengthen it.

[dkyndt]

Thank you for you answer,

By your answer I presume the similarity value is not available in one or another way.
So I'll have to do it the long way by checking all pictures for faulty face assignments.

BR Davy

[Mario]

I don't understand what you want to do with the similarity value...

IMatch calculates a similarity value dynamically for each person when judging which person to assign to a face.
It also uses this measure to produce the "best matches" for the Person Selector, given a specific face.

When you would look at a face, IMatch can tell you the 5 persons most likely for that face. It does so already, in the Person Selector.
It is also used to assign faces to persons. You see the result in the unconfirmed faces for a person (also in the Face Manager).

How would you use this value "in another way"? I don't understand what you try to explain, sorry.