Handling face annotation with a cross

[mastodon]

A have made a lot of face annotations with Picasa. That have been imported by IMatch, but lot of the annotations are crossed. If I modfy it (the face annotation frame) a bit, the AI modify the name on it, but I don't want to change it (most of times it fails). Even worst, the AI will modify the other tags in the picture.
Why this crossed annotations appears? How to transform this crossed annotations to true manual face annotations or to confirmed face annotations?

[Mario]

An annotation with an X usually means that no face fingerprint could be extracted.
Are your files online or do you work with offline files? What does the IMatch log file tell us?

[mastodon]

All files are online, on an other HDD than the systemn and IMatch. Might be the problem, that windows have stopped the HDD during face recognition?
I have made a full log of a changes in one picture (img_6059). And even the faces that have been recognised by the AI, are sometimes crossed. I attached an example.

ADMIN: Image removed. It is illegal to post images of identifiable persons without explicit permission.

[Mario]

This looks normal.
A face rectangle imported, but for this face IMatch can not extract a face fingerprint (too dark, occluded face).

[mastodon]

But it was not me, that was made by the AI. And it did not imported very easily recognisable faces, small but light.
What can I do with this tagged, but not imported faces? I would like to have them as confirmed faces (if AI can make fingerprint) and true manual faces. BUT I see that AI does not recog them good. I want AI to get the fingerprint of the face and import it to the person I have tagged.

ADMIN: Image removed. It is illegal to post images of identifiable persons without explicit permission. Especially images of socker players, where rights are held by UEFA, their clubs, their management or the players themselves. These are not people. These are brands.

[Mario]

I don't understand what you want.
The AI can find the face, but not make a fingerprint. This can happen. Just assign a person to the face. They work like any other face then.
This is also explained int he People help topic.

Have you really created these face annotations in IMatch? These look like imported XMP tags to me...

[mastodon]

Yes, the major problem for me is that imported face tags are not assign to person. Yes, they was made by Picasa, but most of them are clear, easily recognisable faces. So, the AI should have make the fingerprint. And if I modify a bit the frame of the annotation the AI recognise them as a face. Unfortunatelly not always correct, so I have to repair the tag.
So, I had face tagged pictures in Picasa. In IMatch 2019 it was OK. I import them to IMatch 2020, and after that IMatch AI does not recognise them as face, I have to tag them again. Why not to connect these face tags (John Doe) directly to the person named John Doe?

[Mario]

Well, I have tested about 30,000 files with XMP face annotations created by Picasa, Lr and other, more professional software.
IMatch was able to use the face regions in 96.3% of these test cases to produce a face fingerprint.

Unless you images are something very special, IMatch should be able to do the same for your faces.
Feel free to send me an an example image with XMP face regions for which IMatch fails to detect the face to support email address
Also include your face-detection settings, unless you use the default settings.

Please include a link to this community thread. I have 130 emails to process in my inbox so it take a while until I can look into your particular problem.

[mastodon]

I run a test with "Re-create face annotations from XMP data", and that solved the problem. That let import face annotations to true manual face annotations or confirmed face annotations.
It is strange, that the AI could not recognise as a face very good, light and cllear faces from above. But, that is not a bug!
So, at the end it will be OK: I will "Re-create face annotations from XMP data" for every file.
Thank you for your help!

[Mario]

When IMatch 2020 migrates your database, it looks for existing face annotations.
If the file is online or a cache image exists, it uses the bounding box of the face annotation to create a 'real' face and fingerprint, and the annotation label to produce persons.
If you have initially created these annotations by letting IMatch import XMP regions, there should be no difference.

Or did you perhaps use 'invalid' characters in your face annotation labels (like comma) and thus IMatch failed to link them to persons during migration?
See this bug report: https://www.photools.com/community/index.php?topic=9762.msg69028#msg69028 and the corresponding fix included in IMatch 2020.3.2 (https://www.photools.com/release-notes/?productId=372 and there #01006

[mastodon]

One question left, how can I filter pictures, that no face fingerprint could be extracted. It would be quicker to run that command only on files, that have this problem.

[Mario]

This is so rare, there is no feature for it.
This can only happen if you have annotations in your old database which are not enclosing an identifiable face. I have found that to happen in less than 1% of my test cases. So, no special feature for that.
As I said, your images must be real special. Feel free to send me some samples to support email address I can include in my test library.

[mastodon]

OK, if I stuck or feel overhelmed, I will mail you. Thank you.