Lightroom claims IMatch's Keyword List contains corrupted data

[Damit]

I am trying to import my IMatch keywords into LR. However it states that the keyword list contains corrupted data.  I cannot figure out what it is referring to and have tried many different keyword lists generated by IMatch at different times and all give me the same result. They are in the proper UTF-8 format.

However, if I try to import a keyword list from photo-keywords.com it imports to Light Room with no issues.

I have attached the Keyword list in case it helps.

[Mario]

When importing the file, LightRoom tells me:

One or more of the keywords being imported contained illegal characters (such as "|" or ",") that were converted into the "-" character.

[Content]
Problem keywords: Restaurants, Bars, And Pubs, Comets, Meteors, And Asteroids, Congo, Democratic Republic Of (Kinshasa), Congo, Republic Of (Brazzaville), Korea, Democratic People's Rep. (North Korea), Korea, Republic Of (South Korea), Myanmar, Burma, Micronesia, Federal States Of, Macedonia, Rep. Of, Moldova, Republic Of, Congo, Democratic Republic Of (Kinshasa), Congo, Republic Of (Brazzaville), Korea, Democratic People's Rep. (North Korea), Korea, Republic Of (South Korea), Myanmar, Burma, Micronesia, Federal States Of, Macedonia, Rep. Of, Moldova, Republic Of and Miami Dade Community College, South Campus

The file contains no | but you have used commas in several places, and Lightroom does not handle commas when importing.

For example, Lightroom mentions Democratic People's Rep. (North Korea) as a problem and indeed in your file it is listed as

Korea, Democratic People's Rep. (North Korea)

Lightroom does not like the comma after Korea so you must fix the file before importing. Replace , with something else, e.g --COMMA-- and then find these keywords after importing them in Lightroom and replace them again with a comma. Or just let Lr replace them on import and then edit afterwards.

This is a limitation in Lightroom and not a problem caused by IMatch. Characters like , ; and | often have a special meaning for different applications and you should avoid them in keywords.

That being said, Lr Classic 12.1 has imported the keyword list just fine. No message about corruption or anything.

[Damit]

Thank you so much, Mario for the explanation.It is good to know that other applications use these characters for special purposes. I will stray from employing them from now on. It is strange because Lightroom 12.2 will not let me import the list due to corruption as well.  I am surprised that one cannot include a comma, but it is what it is.

I did try your suggestion (--COMMA--) and that imported.  Your advice will allow me to correct things and choose a proper replacement.

I am now realizing that I really need to get disciplined with the keywords in both applications.  I assume I need to just delete the current keyword set in LR, import the Keyword list from IM and then have LR read metadata from files to populate the keywords. I then must make sure to not create any keywords in LR and import any changes to keywords in IM to LR.

I would do everything in IM but face detection is working much better for me in LR than IM, for whatever reason. So I guess I just need to worry about my people Keywords.

[Mario]

I have the latest Lr and it imports your file just fine (automatically replacing the commas with -).

I am now realizing that I really need to get disciplined with the keywords in both applications. 

Maintaining and modifying keyword lists in multiple applications with constant import/export is not a really good workflow. It will fail some day and you forget to export/import...

(...) but face detection is working much better for me in LR than IM, for whatever reason.

Classic or Cloud-based where your images are processed by the Adobe AI in the cloud and not locally?

Usually people tell me that the face recognition in IMatch beats Lr in most cases.
Did you check the trained faces you have created. Often in such cases the users trained the wrong faces, limiting the recognition of persons in IMatch severely.

[Damit]

You are right. It is not a good idea. I have to figure out how I will do this, but it will have to be manualized and disciplined, if I continue to employ this workflow. I am only using LR for face recognition, so I just need to keep my People synchronized. I have to think about this carefully.

I am using Lr Classic.  I don't believe in cloud based applications.  Probably old school, but that is my position, for now.

I really hate to say it.  Believe me I would love to do everything in IM, but the difference in accuracy and ease of use with Face Recognition is night and day.  I have to correct or create at least 3X more images with IM than with Lr. I have checked trained faces multiple times.  I have tired to get it to work at least half as well, but for whatever reason, in my case and experience, IM is not cutting it for Face Recognition.

Maybe I will have to resort to linking people to resolve these issues. I don't know, but thank you for your help. I will reevaluate my trained faces again. Maybe I am missing something.

[Tveloso]

I really hate to say it.  Believe me I would love to do everything in IM, but the difference in accuracy and ease of use with Face Recognition is night and day.  I have to correct or create at least 3X more images with IM than with Lr. I have checked trained faces multiple times.  I have tired to get it to work at least half as well, but for whatever reason, in my case and experience, IM is not cutting it for Face Recognition.

You might also want to check if you have lots of Rejected Faces for the persons not being accurately recognized/assigned.  Removing the rejected faces may improve the accuracy.

For me, Face Recognition in IMatch is very nearly 100% accurate (both for the 10 or so people that frequently appear in newly indexed files, and for those that appear less often).

But in the early days of Face Recognition in IMatch, I had similar, very poor hit rates (particularly with my wife and daughter, who resemble each other).  When IMatch would confuse my wife and daughter, I would reject the unconfirmed face assignment (not this person).  After a while, IMatch not only stopped confusing them, but also stopped assigning the correct person to both of their faces (assigning instead all manner of other persons in my database).

Another user here suggested that rejecting faces might negatively impact future face assignments, because IMatch was using rejected faces as the basis to not assign other similar faces for the person (including probably that person's own face).  The thinking was this:

1.) IMatch created an unconfirmed person assignment for a face (which was not the correct person)

2.) That assignment is rejected by the user

3.) IMatch will now be less likely to assign the person's actual face, since it's similar to the rejected face (that's why it was originally suggested in #1 above).  So Rejected Faces acted as sort of Trained Faces for "not this person".

So I "deleted" all my Rejected Faces, and afterwards, my wife and daughter were now correctly assigned almost always.  IMatch occasionally still confuses them (but rarely, and understandably so).

To "delete" the rejected faces, in Face Manager select the Rejected group from the dropdown, select all faces, and click the Reject button (to toggle the rejected state back off for those faces, for that person) - do not click Delete (which will delete the Face Annotation from the file).

I'm not sure if this is actually still needed, as there have been many improvements in Face Recognition in IMatch since then, but I have continued to avoid creating Rejected Faces, and the recognition rate is extremely high for me.

The only place where I think there might possibly still be a small issue is in Face Detection during the "batch Face Recognition operation" in the File Window.  Occasionally, while reviewing newly created Face Annotations in the Viewer, to manually place Face Annotations (on the backs of people's heads, etc), I will hit upon a file that doesn't have any Annotations, which I believe should have them.  Pressing F6 in the Viewer, creates those missing Annotations (most times already in a confirmed state, with the correct person assigned)...so the "batch Face Recognition operation" in the File Window seems to have skipped processing that file for some reason.

[Mario]

Rejecting too many faces can actually bring down recognition rate, especially if you have persons with very similar faces (siblings, parents and children, persons covered over a long period of time).

The IMatch help explicitly mentions this nearby the reject command:

Instead of rejecting a face, it is often better to assign the correct person. Or, if the person is unknown, to either ignore or delete the face.
Rejecting too many faces for a person may result in too many (correct) faces not being considered as matches for the person.

[Damit]

I follow the advice in the Help section.  I do assign the correct person and I usually delete faces I do not use.  I will try and try again, to get the FR to work better. Like I said I would like to do all my organizing work within IMatch.

Quick question: If you add a person, does it add an entry to the Thesaurus. I think not, but I want to confirm. I know we can add keywords to People but I am wondering if IMatch does this automatically for the named person. Again, I think not, given Mario proclivity to complete user control.

[Mario]

IMatch optionally ads new keyword added in the Keywords Panel to the Thesaurus (this is a setting in the Keywords Panel).
You can also manually add keywords or other tag values to the Thesaurus using the corresponding commands available in the Metadata and Keywords Panel.

Adding a person does not change the Thesaurus, else it would be documented in the People help topic: Person Keywords and Categories