Filtered Face detection, using & converting links to annotations

[Damit]

It would be great if you could tell IMatch which faces to use when detecting faces. For example, you may have a trip that you took with your wife and child. Telling IMatch to detect only those haves should help avoid errors and streamline the process. IMatch would ignore all faces that did not meet the parameters of the defined face set.

In addition, it would be great if you could tell IMatch to use linked people's names when doing face detection, and then convert the link to a real face annotation.I do find it odd that when I move a face link to a face and convert it to a real face annotation, that it does not at least suggest the person linked.

[Mario]

In addition, it would be great if you could tell IMatch to use linked people's names when doing face detection, and then convert the link to a real face annotation.I do find it odd that when I move a face link to a face and convert it to a real face annotation, that it does not at least suggest the person linked.

If you convert a link to a normal face annotation, it does exactly that. The face annotation is converted into a manual face annotation and the person stays assigned.

But when you now move and/or resize the annotation (I assume this is what you do), the regular face annotation processing is triggered. After you finished resizing / moving the annotation, IMatch tries to detect a face inside the area and, if it finds one, assigns the best matching person.
If the same person is detected, it remains assigned. If the AI considers another person more likely, that person is assigned. If no face is detected in the new position, the existing person remains assigned (from the link) and the face annotation remains in manual state.

If it would just assign the person of the initial link this could flood the on the facial fingerprints the AI maintains in the database with 'bad' data, reducing the overall quality of future face detection. Much like training the wrong faces or too little variation of faces has a negative impact on overall face recognition quality.

If you run into problems like this and for your particular workflow, I would suggest you save the step of adding a face link first. Instead, create a face annotation with <Ins>, position it and assign the person. Use copy & paste to copy the annotation to other files and add the person to the gallery to save time in the future.
 
Face links are just a crutch to help users which don't want to do face recognition or which have too many unrecognizable faces in their collection. Links integrate these special cases in the IMatch 'People' feature area without the need to do real face annotations.

Usually you set the link and leave it be. Else it is more effective to manually create face annotations and directly assign the person. In case the IMatch face recognition does not work. Which is rare, but can happen. No AI is perfect, not even the ones from Google or Facebook or Microsoft.

[Damit]

Thank you Mario.  Yes, I am immediately moving and resizing it, which explains why I did not even realize that the name initially stays the same. It would be great if you could add an option to not have a face link trigger regular face annotation processing, but I don't know if that may complicate something else I am not considering.

Your work around is an interesting proposition. To make sure I understand correctly, are you saying I can copy a face annotation from one file and then paste it to multiple files?  I assume the coordinates would remain the same, would that not create problems with many face regions not containing faces? Still, it is an option I may consider.  One way linking people would help is that I will know which files need to be worked on while using your workaround would require me to somehow keep track of the files I need to appropriately face tag, probably with a category.

I am linking people instead of using face annotation on some files for the exact reasons you stated.  Sometimes I need to move quickly and it is much quicker to pick a bunch of files and link a person than to go through each file and create face annotations.  I am doing so with the idea of going back and changing them to face annotations, if I have the time.

I also have too many unrecognizable faces and am not found of ignoring faces.  I usually delete them because I find the ignored faces just clutter up my Unknown Persons. Since I usually will not run face detection again on most files, I assume there is no negative consequences in just deleting the face regions. As you know, I am still relatively new to all this, so maybe my approach is flawed.

Still, I think my suggestion of a filtered face recognition would be useful. And I believe using linked regions would also present a valuable option in this endeavour. Unfortunately, I am not having good results with face recognition, either here or in other applications. Amazon photos was the best I tried, and was actually very good, but I do not know if it writes the face regions, plus I am concerned about privacy issues.

[Mario]

Your work around is an interesting proposition. To make sure I understand correctly, are you saying I can copy a face annotation from one file and then paste it to multiple files? 

Copy & Paste Face Annotations

User Annotation Gallery

[Damit]

Thank you! This is very interesting.  One problem I have is that the Annotation Gallery is very small, so when I add the faces to the gallery, I can barely see them, not do I see a name below it, only when I hover over it, so it is not really useful for me. I will play with copying the face regions and see how it goes.

[vsammy]

It would be great if you could tell IMatch which faces to use when detecting faces. For example, you may have a trip that you took with your wife and child. Telling IMatch to detect only those haves should help avoid errors and streamline the process. IMatch would ignore all faces that did not meet the parameters of the defined face set.

That would be very helpful, 70% of all my pictures taken are within (my big) family.

[Mario]

I don't think this is needed. The FR in IMatch is very good and false positives (wrong face pickled) is usually caused by wrong training (check the trained faces) or super-similar persons like twins, siblings or small children. And in these situations limiting the persons to use for face annotations would not help, most likely.