Manage Faces in XMP regions options missing from Metadata 2 (Lightroom to IMatch)

[Damit]

I was reading here: https://www.photools.com/help/imatch/rmh_config_metadata2.htm on how to work with Lightroom and IMatch together. In these instructions, there was a mention of a setting for "Manage faces in XMP regions." I cannot find this setting. 

Currently I have done some testing and it seems the Hierarchial Keyword which is populated when I draw and name a face region in Lightroom is updated almost immediately, but the face region and the Person in Image metadata tag does not seem to transfer. Maybe that is the setting I have to adjust that is missing.

I would like to use Lightroom's face detection with IMatch so if any other information, such as contingencies, startegies or instructions on how to do this properly exists, please point me to it. Thanks!

[Damit]

Well, I figured out that the reason I was not seeing the option was because I had to be in Expert mode.  The instructions briefly mention it, but do not tell you how to do this (It is in Edit > Preferences > Application...). The only expert mode I found in the Help Manual was in the dashboard and I enabled it there but that did not reveal the option that was missing.

Anyhow, the option for "Manage faces in XMP regions" is set to yes and yet the face annotation region and name (person in photo) do not transfer from Lightroom to IMatch.  I suspect it may be the "Keep existing XMP" setting, which is currently set to yes.

Again, any tips for keeping thing straight between the two programs would be appreciated.  One thing I wonder is if I should Synchronize the folders in Lightroom before making any changes there.  I suspect yes, but I try to assume as little as possible.

[Mario]

The XMP regions transfer just fine from Lightroom to IMatch. I use Lr myself.

Keep in mind that you must explicitly make Lr write out XMP data. Else it is just kept in the proprietary Lr catalog, inaccessible for IMatch and other software. This is how Adobe prefers it. See the Lr manual for details.

You can easily check if the images contain face regions in the IMatch ExifTool Command Processor
("List Metadata", then search for region) or just switch the Metadata Panel to Browser mode and look for regions.
If there are regions and they are not malformed, IMatch will import them. This is the default and it usually just works.

The Expert Mode is intentionally hidden because users should not find it accidentally.
It is for experts, and the options revealed by turning it on can cause problems or even data loss when used wrongly. And users frequently tend to do weird stuff, shot themselves in the foot and then cause support cost to resolve the problems they have created. This is why I have hidden some features, commands and options behind the Expert Mode option. It's just better that way.

The defaults in IMatch work just fine for 98% of the user base. Face regions written by Lr, Picasa etc. are important automatically.
I suggest you turn the Expert Mode off again unless you really need it. This may prevent you from making a wrong setting, accidentally causing trouble now or later.

Make sure Lr writes the XMP to the image file, and IMatch will pick the changes done in Lr automatically.
Unless you a) have modified face regions in IMatch and b) have metadata protection on. In that case, IMatch must not replace XMP data that has not been written when ingesting a file. That's the purpose of metadata protection.

Write XMP data in Lr, let IMatch pick it up.
Let IMatch write-back, then re-import the files in Lr with the icon. Lr does not automatically reload metadata written by other applications. Depending on your version you will get a prompt about modified metadata and if you want to re-import it into your Lr catalog.

[Damit]

I have set the Catalog settings to Automatically write changes to XMP, so that should have LR writing the face regions. I looked in the browser metadata and I see the initial face region that existed before I used IMatch and which imported to IMatch when I initially imported the picture, but the region I made to test how things transfer is not there. However, the value I gave to this region "test" does show up in the keywords in the XMP Adobe Lightroom>Keywords and the IPTC ApplicationRecord>Keywords sections. So, something is transferring, just not the face region with corresponding annotation.

Apparently, you edited your response, so I am editing mine

I will turn expert mode off.  I tend to get myself in trouble, so I will take that advice.  However, I may need to change something in there because inevitably I will end up working with a file I had previously worked on with IMatch.  I did work on over 30K faces in IMatch and I may want to add more faces from those photos I had not previously added in LR before I initially imported them into IMatch. I wish I could do everything in IMatch (IM), but LR works better for me for face recognition, probably because I am doing something wrong in IM. Still, I like the LR user interface with the dragging of faces and I want to see if Sensei actually learns better than IM's facial recognition.

Anyhow, since I have such a large group of files that were edited in IM and that I will work with these files again in LR before bringing them back to IMatch, then I believe I have to turn "Protect existing XMP" to "No."  In order to protect existing XMP I have changed in IMatch, I believe I need to reload metadata written by IMatch in LR.  I thought I did that by synchronizing the folders in LR but that assumption is probably incorrect and even worse, may be problematic.

I have the latest version of LR Classic but have never noticed a prompt or icon about modified metadata.  Can you point out where that is?

Lastly, I want to that you for your help! It is very much appreciated. I am trying to do my best to figure things out on my own.

[Mario]

If the region is shown in the MD panel but not in the Viewer or Quick View panel, make sure the display of annotations is enabled. That is the Pause button in the Anno Panel in the Viewer. 

 

Unless the region is broken somehow (then IMatch would have logged it into the log file) this is the only reason I can think of why it would not show in the Viewer if it has been imported.
Attach your test file as a ZIP so I can see the region and analyze why IMatch would not import it.

Something similar was never reported before.
Except for some face regions created by Apple devices, but these regions had both width and height set to 0 (zero).
IMatch already has a work-around for that.

[Damit]

The region and the annotation does not show in the IM browser, only the keyword (in XMP Adobe Lightroom section, both in Hierarchial keywords and Weighted Flat Subject & Also in the IPTC Application record) that it populated in photoshop.  So, something is transferring, just not the face region or annotation.  Again, I think it is that Keep existing XMP setting.

Here is an image that may make things clearer:

[Damit]

Well, that did not work.  Here is the link to the pic:

https://imgur.com/a/ORdag7Z

[Mario]

This is a screen shot of the Metadata in IMatch, which does not help much.
I need the image you produced in Lr (the one you import into IMatch). If I get a face region here, we know it is a setting or something else on your PC. If I get no face region, I can look at the data Lr has written to figure out what the problem is.

You can upload the image to your cloud space (OneDrive, GoogleDrive, DropBox, ...) and send a link to support email address with a link back to this thread (!).

NOTE: Please do not force me or other users to download images or screen shots from 3rd party source.
You can attach image files directly to your posts in the "Attachments and other options" section below the post editor.

The site you uploaded the image to sets cookies from 44 (!) other web sites, including Facebook and Google ad networks. Sending my IP address to dozens or even hundreds of 3rd party data collectors.
Just to look at a screen shot from your computer.

[Damit]

First, sorry about the 3rd part source.
I tried to attach directly using the icon and saw a URL, so I thought I had to do it that way. Thinking back, I could have used a link from my One Drive, as I just did in the email I sent to you. Now that I know about that drop down, I will attach there from now on.

I posted that image to show that the name of the Face Annotation "test" was present and transferred, in the IPTC ApplicationRecord and in the SMP Adobe Lightroom sections, but the NWG Regions did not. They only showed that which was defined in IM.

I retagged the photo again in LR (Which is the state of the pic that I sent in the email) and IMatch now shows two sets of MWG "Face" Regions, with two sets of coordinates, separated by semicolons. However, the viewer does not show the box or the annotation. I included a screen shot of the metadata panel and adjacent to it, the viewer image of this file, in case it may help.

There seems to be something cancelling out the other because as I wrote metadata changes to IMatch, the two face regions disappeared leaving just one: Jorge Salazar Carillo, so somehow IMatch overrode the Region changes done in LR, but not the keywords. I just tested this out again and the same thing occurred. I clicked on the little pencil, wrote the metadata changes and the Region changes in lightroom that were shown in IMatch prior to writing the metadata reverted to just one person. It is like it just reverted to the state prior to the changes in LR. I am sending you an email with the file in the state after updating the metadata, in case that helps.

This is something I need to sort out because this can cause pretty bad results. I need to ensure that the changes I make LR stick and that I am doing things correctly before I proceed.

[Mario]

Send the sample image as I asked for above. so I can check what metadata it contains. All else is guesswork and a waste of time.

This usually just works and many IMatch users use Lr. Including me.

Keep in mind that editing metadata in multiple applications will cause issues.
Adobe did not design XMP to be collaborative as a major design goal.

Modifying XMP in multiple applications simultaneously requires a lot of discipline. Lr writes metadata asynchronously and may sometimes delay it. IMatch has metadata protection which may block updates when you change the same tags in IMatch and Lr.

Always ensure to write back in IMatch and re-import the XMP into Lr before you process an image - else Lr will wipe out the changes done by IMatch.

Giving the excellent quality of the IMatch face AI, there is usually no need or advantage in adding face annotations in Lr. Same for metadata.

[Damit]

I had sent the image prior to writing my response and I just sent an email with the file after I wrote the metadata changes in IM, which seems to override the face regions. Have you not received them?

Again, I want to point out that even when IM showed the two face regions in the metadata panel, it still did not show the box or the annotation in the viewer.

[Damit]

I have experimented several times with this.  I have changed each of the "Protect existing XMP" and "Keep existing XMP" to yes by themselves and also together.  Each time I make a change in LR and save the metadata to the file (I am aware that you must do this instead of assuming LR does it immediately), IM updates the file rather quickly by itself and reflects the changes in the MWG Regions metadata but does not actually produce the face region or annotation in the viewer.

IM then adds the pencil icon to the file, reflecting that the file has unwritten metadata. The only change I made to the file was in LR, so this must be the metadata IM has flagged as changed. However, when I hover over the pencil it lists the tags that need to be written as IPT::ApplicationRecord\Keywords & XMP::dc\Subject, which is curious because it is not listing the MWG regions.

When I click on the pencil to write the metadata changes, the IPT::ApplicationRecord\Keywords & XMP::dc\Subject actually do not change at all, only the MWG Regions.

[Mario]

I have downloaded your sample files.

When I import the image, IMatch shows two face annotations. One for the elder gentleman (Jorge) and one labeled "test" which is not a real face. I see the same annotations in Lr.



This is what I see (I blurred the image for privacy reasons).

Now I add a 3rd face annotation (for the pepperoni), add a rating, label, title, description, keywords etc and write back.
Lr picks up the changes when I click the reload metadata icon and now shows also 3 face annotations:



This is exactly as it should behave. Please make sure you have annotations enabled in the Viewer, than you use the standard Metadata Settings in IMatch and that Lr has reloaded the metadata written by IMatch (it shows a small icon on the thumbnail otherwise). Make sure Lr has written metadata to the file (IMatch will automatically reload the file afterwards).

I can see no problem. I work with Ps, Lr and other applications and the metadata written by IMatch always shows up (depending on the metadata capabilities of the application) and IMatch always picks up changes done to metadata by other software.
 

[Damit]

I tired your test and it works, which means LR is detecting and representing changes made in IM accurately, which is great, but it is not the problem I have described.  It is the other way around.  I need IM to accurately represent changes made in LR. The problem seems to reside in IM.

Please run the test this way:
Add an annotation and face region in LR. For my test, I added a field and annotation around the hand holding the hat and called it "handy." After defining and annotating the region in LR, make LR write back the metadata. 

Now go to IM and, after a minute or so, you should see the metadata representing the regions and the annotation in the browser. Now open the file in viewer. You should notice what I am noticing, that neither the face region nor the annotation for "handy" appear in the viewer, as they should.  The metadata is there, listed in the MWG Regions listing in the browser, but IM is not drawing the box or providing an annotation, presumably because there is no graphic representation of a box.

That is an issue right there.  The box and annotation should appear according to what I have read about the capabilities of IM.

Now go back to the media and Folders and click on the image. You will see, again, in the browser that all the metadata is present for all region and annotation for all four regions: handy, Pep, Jorge Salazar Carrillo and test are in the MWG Region section with 4 sets of region coordinates and dimensions. You will also notice the pencil icon on the image. Process the metadata changes in IM.

You will notice that the region and annotation metadata under MWG Regions for "handy" added in LR will disappear, leaving only those you had previously defined in IM. You will also notice that "handy" is still present in the XMP Adobe Lightroom and IPTC ApplicationRecord fields. That is the second problem. 

I do not understand 1.) why the metadata for XMP Adobe Lightroom and IPTC ApplicationRecord fields remain, while the others in the MWG region are erased and 2.) why the recognition box and annotation are not appearing in IM when they are showing in the metadata fields of IM and the boxes do show in LR where they are defined, but that is what is happening repeatedly.

Please run the test this way because your test was not representative of the problem I described. Hopefully you will notice the issue, which hopefully can be corrected. Thank you!

[Mario]

Ah. Well, this explains your problem.

When a file in IMatch already has face annotations, IMatch does not unconditionally wipe them and replace them when metadata is re-imported because a file has been modified externally.

If you have images with face annotations in IMatch and you then add additional face annotations or modify XMP face regions in an external application (which you should not do, obviously) you need to:

a) delete all face annotations in IMatch (Viewer)
b) force a reload of the metadata via Shift+Ctrl+F5

IMatch prioritizes its internal face annotations, because they are so much more than simple XMP face regions. Wiping them accidentally could be potentially disastrous.

I recommend to settle for one application to create face annotations. Either use Lr or IMatch. Or whatever. Don't modify metadata in multiple applications, because this very likely will create problems.

When you really, really must must modify (why?) face annotations for images already in your IMatch database in Lr, perform the steps explained above. See also the IMatch help: Using Google Picasa™ / Adobe Lightroom® Face Tags in IMatch which explains this.

But this is such a cumbersome workflow, I cannot really recommend it. Unnecessary work.

Stick to one application for doing face recognition. If you prefer Lr, do it before importing the files into IMatch.
Please understand that IMatch cannot support or cater for all possible workflows users come up with.
Since you are AFAIK the first user ever who did run into this 'problem', I assume that most users let IMatch do it's thing or import face regions only once and then modify them only in IMatch.

Ps.: When I add more face annotations in Lr and do the a) b) dance described above, IMatch of course will show the face annotations exported by Lr into XMP. And it will do all the work again already done before the existing face annotations were deleted. Extract face data, try to assign persons etc.

[Damit]

Yay! At least I know now why this is occurring!

Please understand that IMatch cannot support or cater for all possible workflows users come up with.

Of course not! I would not expect this. But this was stated in the help article I posted in my initial post about the Manage faces in XMP regions option: "If this option is enabled (default), IMatch imports face regions from XMP into IMatch face annotations. These face regions may be created by applications like Google Picasa™, Adobe Lightroom® or other software with face recognition technology. This option also controls if IMatch stores face annotations back into XMP during write-back," so that is why I asked the question.

Now I know that it does, as long as you have not done any face recognition through IMatch. I highly suggest you add a link to the help article you suggested (https://www.photools.com/help/imatch/people.htm?dl=h-116) in the help article I read (https://www.photools.com/help/imatch/rmh_config_metadata2.htm) under the "Manage faces in XMP regions" section, because it would have made me (and others in the future) aware of this caveat, the resulting dance and how to approach the work. I must say that if no one had had this issue, you would not have written the article, so I do not think I am singular in experiencing this "issue." Still it is what it is, and is not a deal breaker for me. I am really exicited about the other possibilities of this program.

I do wish IM would let you import additional and perhaps override the present face maps because the same thing happened with Picasa, though I was not so meticulous in my tests. Maybe I will put it in as a feature request and see if others weigh in.

I also did work on over 35k of face tags due to my ignorance of this "issue, " and feared that it may have been a waste of time. Luckily I think I can overcome this. As a work around, I can just delete all the face tags in IM AFTER I have LR read the metadata or synchronize the folder (same thing?). I did not notice the icon that shows the file has changed metadata in LR on the file I had just changed, but did notice it in another.

Unfortunately, in my experience thus far, IM is not doing a good job with face recognition, I don't know why, but the performance I am experiencing is poor, especially when I add manual face detection, where updating the person can take long periods of time. I am not sure why. I have tested the face recognition capabilities in the same folder in LR, Picasa and IM and, for me, IM had the worst perfomance.  It was not close. Even now, with all the face I have detected and taken great care to chose good faces to train with, it makes some pretty bad suggestions on newly detected faces. I really wish I could report otherwise.

Ps.: When I add more face annotations in Lr and do the a) b) dance described above, IMatch of course will show the face annotations exported by Lr into XMP. And it will do all the work again already done before the existing face annotations were deleted. Extract face data, try to assign persons etc.

This is good news except for the last sentence.  Will IM ask me to confirm all the faces I assigned in LR?

[Mario]

Unfortunately, in my experience thus far, IM is not doing a good job with face recognition

Interesting.
As far as I'm aware, the face recognition in IMatch is considered to be one of the best available systems on the market. Excellent recognition rate, no cloud, no privacy issues. User-configurable etc.
Many people out there seem to be very happy with it.
Your personal mileage may vary, though. Use whatever works best for you.

Keep in mind that I neither have the budgets of Adobe or Google or Microsoft or the 50+ persons team size when it comes to face recognition or AI. I'm sure they may be able to develop far better AI solutions than I can...?

If you find something better than what's available in IMatch, go ahead and use it. IMatch does not lock you in.

Just follow the a) b) rules from the help I mentioned above when you change face regions for images already imported into IMatch and things will work the way you think they should work. Yours is a very particular workflow, but IMatch still has you covered if you follow the instructions.

I highly suggest you add a link to the help article you suggested (https://www.photools.com/help/imatch/people.htm?dl=h-116) in the help article I read (https://www.photools.com/help/imatch/rmh_config_metadata2.htm) under the "Manage faces in XMP regions" section

I don't think linking too many help topics to too many other will help at all.

The "How IMatch deals with XMP face regions produced by other software" is an intrinsic part of the People help topic, and that's where it belongs to and actually is. You have just missed reading or following it. Can happen. No harm done.

Who looks at the by default hidden expert mode settings before reading about how IMatch works with face regions created by other software? I estimate a very small portion of the user base.

If you think something is missing in the help, use the feedback link available at the bottom of each help page to report missing, invalid or incomplete information. Feedback emails sent using this function are automatically pipelined and I look at them every couple of weeks to see where and how I can improve the IMatch help.

This is good news except for the last sentence.  Will IM ask me to confirm all the faces I assigned in LR?

Have you tried it? I think this is explained fairly well in the People help. Make sure your tags match and IMatch it will not have to ask any questions.

[Damit]

Who looks at the by default hidden expert mode settings before reading about how IMatch works with face regions created by other software? I estimate a very small portion of the user base.

LOL! Yep, that would be typical of me, I tend to be the outlier.  As I said before, I do tend to get myself in trouble. When I scrolled up and realized where that help section on Picasa and Lightroom came from, The People section, I could not believe I had not read it. As aforementioned, I do try to read and not ask stupid questions, though my efforts seem to falter, as they did here. Thanks for your understanding. I believe I got to the section or two right before it and then I was inspired to read some other topic on IMatch and must have got sidetracked before reading the whole section and forgotten to finish it. 
I do think since the Metadata2 topic specifically mentions Picasa and LR that there is a strong association and the link would be helpful, but I understand your position. I would report it in the feedback, but since you already considered it and decided against it, there is no need.

Keep in mind that I neither have the budgets of Adobe or Google or Microsoft or the 50+ persons team size when it comes to face recognition or AI. I'm sure they may be able to develop far better AI solutions than I can...?

That is humble of you to say, and it may be true, but is not always the case. Sole developers have more freedom and imagination which sometimes can bring innovation others are too bogged down with politics to enact. I am sure your AI is good, as you said, YMMV, but I am surprised that your face recognition is not doing better. I must reconsider the trained faces, but most of the faces originally came from LR during transfer. Maybe that has something to do with it?

Believe me, I really want to do everything in IMatch, but unfortunately it does not seem to be the best solution for me now, as far as facial recognition.  However, I did not find or come to IMatch for facial recognition; there is so much value in the other functions of the program that it merits a little bit of work-flow complication, if necessary, to use it. After retrying all the facial recognition options I have, between ACDSee, LR, Picasa and yours, LR does the best job and I like their user interface and workflow, with the dragging of faces, so I will stick with that, for now.

Frankly, I have only dabbled with face detection as I have really put off organizing my photos because I have been strategizing on the execution for years. I have tried Amazon, Google, LR and ACDSee in somewhat extensive trials with varying results.  Cloud systems like Amazon performed well, but I had privacy concerns, and I don't know how to export their face regions to something LR or IMatch can use. I have put a lot of thought and planning to building a protocol. I have read many articles, forums, the DAM Book, and your help sections, which I have found very helpful, as they are replete with great ideas and tips. I think I am nearing a protocol to begin the task in earnest.

Just follow the a) b) rules from the help I mentioned above when you change face regions for images already imported into IMatch and things will work the way you think they should work. Yours is a very particular workflow, but IMatch still has you covered if you follow the instructions.

So, where does the command "Re-create Face Annotations from XMP data" factor in all of this?
I did not realize this existed until scrolling up from your suggested tip.  Why would one delete the faces and force a reload rather than just using this command?  What is the difference?
Also, you mentioned to "a) delete all face annotations in IMatch (Viewer)." Does it have to be done in the viewer?  I thought I could just highlight a folder or all the files with in it in Media & Folders view and then right click and click delete faces.  Can I do that instead?
Still, I assume the "Re-create Face Annotations from XMP data" would perform the same steps as the a) b) dance, does it not?

One strategy I am considering is doing face annotations in LR or elsewhere and just linking persons to files in IMatch. I could use the keywords created when I make the face regions in LR to quickly group people to files and link them. In this approach I would like IMatch to not consider face regions or annotations at all (hopefully until a later date where I find Imatch's face recognition better and import them). I assume by "Manage faces in XMP regions" in Metatdata2 will do this, and I should set it to "no." Is there is another setting to tell IMatch not to import or deal with face regions and annotations that I should know about? Are there other issues I must consider?

I assume Person Keywords and Categories would still apply and I can always copy cover images in a separate folder so I could have a face instead of a placeholder image for my People.

Have you tried it? I think this is explained fairly well in the People help. Make sure your tags match and IMatch it will not have to ask any questions.

OK, I will experiment.

[Mario]

If you don't get good results, check the faces used for training.
IMatch by default uses the 10 first faces assigned to a person, which works well. But if you have faces of wide different ages of a person etc, you did not specify a birth date or the trained faces are not good, to little or too many, uses different faces for training will help lots. 
If the faces in your images are very small and are not detected, use the optimize for small faces.
There are many related topics, tips etc. in the People help topic.

I get a lot of feedback about the very good recognition rate for both faces and persons in IMatch.
There is never 100% but IMatch definitely plays in the Top-5 % according to the benchmarks.

[Damit]

I am trying to get it to work better, but not with great success. I will keep trying, perhaps making a copy and erasing all face data and starting from scratch, but for now I will go with LR for now, while I work on the faces.

As such, if you would be so kind to please address my questions so I can proceed, it would be greatly appreciated:

Where does the command "Re-create Face Annotations from XMP data" factor in all of this? I did not realize this existed until scrolling up from your suggested tip.  Why would one delete the faces and force a reload rather than just using this command? I assume the "Re-create Face Annotations from XMP data" would perform the same steps as the a) b) dance, does it not? What is the difference?

You instructed me to "a) delete all face annotations in IMatch (Viewer)." Does it have to be done in the viewer?  I thought I could just highlight a folder or all the files with in it in Media & Folders view and then right click and click delete faces.  Can I do that instead?

One strategy I am considering is doing face annotations in LR or elsewhere and just linking persons to files in IMatch. I could use the keywords created when I make the face regions in LR to quickly group people to files and link them. In this approach I would like IMatch to not consider face regions or annotations at all (hopefully until a later date where I find Imatch's face recognition better and import them). I assume by "Manage faces in XMP regions" in Metatdata2 will do this, and I should set it to "no." Is there is another setting to tell IMatch not to import or deal with face regions and annotations that I should know about? Are there other issues I must consider?

I assume Person Keywords and Categories would still apply and I can always copy cover images in a separate folder so I could have a face instead of a placeholder image for my People.

[Mario]

Re-create Face Annotations from XMP data

This command deletes all existing face annotations and then creates them anew, from the XMP face regions contained in the image.

This command may be useful if you have created face annotations in an external software, and you want to replace the face annotations IMatch maintains for these files with the externally created face data.

From the help.

If you want to use Lr, this would be one way to go.
Or delete all face annotations in the Viewer and re-import metadata.
or delete all face annotations in the File Window and re-import metadata.
Should all to the same.

IMatch has so many commands and features, I don't recall each and every of them all the time while writing a response to an insanely long thread in the community, sorry.

I have never used Lr for face recog, because I get as good or better results in IMatch.

[Damit]

Thank you! It seemed they would do the same, so I will use the "Re-create Face Annotations from XMP data"

IMatch has so many commands and features, I don't recall each and every of them all the time while writing a response to an insanely long thread in the community, sorry.

No problem, I appreciate the help.

So, if I will just link people in IM for now, should I set "Manage faces in XMP regions" in Metatdata2 to "no?" I assume this will stop any facial recognition and over-riding of changes made in external programs to face regions and annotations.

P.S. By the way, I used the Re-create Face Annotations from XMP data after changing some faces in LR as instructed and I am happy to report that everything worked as it should.

[Mario]

I suggest to keep this option on, since it is the default.
Then decide whether on which program you want to use for fr.
If it's Lr, see my post above. Else this setting will not have any effect anyway, since IMatch does not use XMP face regions when it already has face annotations - see my post above.