Facial recognition in Lightroom utilised in PM

[judophotos]

I don't use Lightroom very much but as I have it as part of my Adobe package I was trying out the facial recognition function and I find that quite interesting. Can anybody tell me where the name of the person recognised and named is held in the metadata in PM?

I have just experimented with a LR catalogue of 4663 images from last weekend and it has singled out over 5300 faces.  Obviously a substantial amount of those are useless and most are probably from the audience but an automated method of identifying competitors would be of great assistance. However, that would only really be helpful if the names could be easily be seen in the metadata and used in captioning etc.

I am using Win10 and PM 5 build 16960.

Thanks in advance for any help and guidance on this.
David

[Kirk Baker]

David,

I don't use Lightroom very much but as I have it as part of my Adobe package I was trying out the facial recognition function and I find that quite interesting. Can anybody tell me where the name of the person recognised and named is held in the metadata in PM?

I have just experimented with a LR catalogue of 4663 images from last weekend and it has singled out over 5300 faces.  Obviously a substantial amount of those are useless and most are probably from the audience but an automated method of identifying competitors would be of great assistance. However, that would only really be helpful if the names could be easily be seen in the metadata and used in captioning etc.

It is quite possible that the identified persons metadata is only held in the catalog database itself and is never written to the XMP.  If you'll post a sample JPEG (that has had its metadata updated) or an updated XMP sidecar file, I'll take a look and see what I find.

-Kirk

[judophotos]

Thanks Kirk.

That is kind of you.

I avoid xmp sidecars so I'll find a jpeg that includes the facial recognition name and upload it.

Best rgds,
David

[judophotos]

Kirk,

A lo-res jpeg from a NEF is attached showing Martyna Trajdos of Germany at a Paris competition on the 6th February. When looked at in Lightroom the name is shown at the bottom of the preview. The NEF image also shows the name at the bottom of the preview.

I look forward to reading your research.

Best regards,
David

[Kirk Baker]

David,

Unfortunately the only time Martyna Trajdos appears in the XMP is in the caption.  I suppose you could ask Adobe how to put that data in the "Persons in Scene" field and then PM and other XMP-savvy applications would have access to the data.

-Kirk

[judophotos]

Thanks Kirk.

If the name doesn't travel with the image file to my Daminion catalogue then I see no reason to use Lightroom.

Thanks again for your help.
David

[soren steffen]

Apparently Lightroom only saves the Persons keyword in the catalog.
I found a bit quirky workaround.
You can create a keyword tag inside the persons keyword and if you uncheck the person box it will show up in PhotoMechanic as a keyword, not as a persons shown in the IPTC.

It is a bit tedious unless you work with the same persons, I presume the judokas are the same in some tournaments? 

I can see it as a help for sports photographers where you can have whole teams and then create the tags inside once.

HTH
Søren

[Linwood]

I can see it as a help for sports photographers where you can have whole teams and then create the tags inside once.

Good luck with facial recognition, I tried it when it came out and ended up just giving up.  It has way too many false positives (identifies person X as person Y) and negatives (fails to see a face at all).  And once it has run on a photo it's just done with it, there's no way to go back (let's say if they improve the algorithm) and have it redo it again for that photo.

The killer for me was the false negatives -- where it just completely skipped photos entirely, so there's no real opportunity to go in and fix it (yes, you can go back and manually mark and tag faces but only if you realize it skipped photos).

Adobe did this and got it out fast for the "me too" factor, but I have yet to find anyone that really got into using it for any volume of work and kept with it.  Maybe you'll be the one, but...

And as far as I know it's never been improved since release, which for a V1 product tells you they either got it perfect, or just lost interest.  I know where my money is.