Lightroom face recognition workflow?

Started by jch2103, August 08, 2016, 01:05:47 AM

Previous topic - Next topic


After seeing several recent posts about using Lightroom face recognition, I decided to try it myself. In general, it seems to work fairly well (does fairly well with full-on faces, not so well with angled faces, and sometimes comes up with very peculiar 'matches' that aren't even faces...).

My issue is figuring out a reasonable workflow with Lightroom and IMatch. I've avoided Lightroom in the past because I haven't ever liked the Catalog and I use DxO for raw processing, hence I'm not really familiar with some of the nuances of how best to use Lightroom, at least in conjunction with IMatch.

I know that I need to have IMatch closed when I'm working on images in Lightroom. I believe I also understand that I need to force Lightroom to write the face recognition metadata back to the files (or xml sidecar files for raw images). However, I'm having some issues getting this work work correctly (I think from the Lightroom end, based on what I see in IMatch (after forcing file scans) and in ExifToolGUI). Only some of my face images seem to get the face recognition metadata properly saved to the files (e.g., to some files in a folder, but not others). Also, I want to avoid writing any other Lightroom metadata to my files, especially as I don't use Lightroom to develop my raw files (I use DxO Pro instead).

Can anyone point me to a correct and efficient workflow/Lightroom settings for pulling Lightroom face recognition metadata into IMatch?


QuoteHowever, I'm having some issues getting this work correctly (I think from the Lightroom end, based on what I see in IMatch (after forcing file scans) and in ExifToolGUI).
I have not experienced any problem there. I import the images in LR, do the face recognition and then select all pictures and use "Save metadata to files". After the face tagging I use DXO to develop the raw images. Not sure how you could trouble shoot this, if you remove the pictures out of the catalog and reimport them or just reread the meta data, are the face annotations for all pictures shown in LR?

QuoteAlso, I want to avoid writing any other Lightroom metadata to my files, especially as I don't use Lightroom to develop my raw files (I use DxO Pro instead).
That's a good point. I just assumed that there are no other metadata because I'm not using Lightroom to develop my raw files either. But I maybe wrong here.


You cannot have one without the other.

Lr by default does not write XMP data to your files, it keeps them only in the catalog.
You need to change your catalog settings to force Lr to always write out XMP data. Or you use a workflow where you manually export the XMP data by right-clicking the files and then Metadata > Save Metadata to Files.

Lr then writes the full metadata, including XMP and probably also updates to IPTC and EXIF (Lr does not document this or I did not find it). It's pretty black-boxes, as usual. Face annotations are part of the XMP metadata record, and Lr always writes it completely, including the proprietary Lr development data, history and whatnot. You have to pay special attention to keyword mapping between flat keywords and hierarchical keywords. You have only limited control in Lr over that, and when you use other mapping settings in IMatch (group levels, hidden levels, flattening options under Edit > Preferences > Metadata) you may create opposing mapping rules which cause both applications to map and re-map keywords all the time...

During my recent work on face annotations in IMatch and the cross-application workflow I also noticed that even when I change face annotations in IMatch and write them back, Lr does not always pick them up. I have to actually remove the folder from Lr and re-import it to flush out the old data from the Lr catalog and force an update.

You don't need to close IMatch or Lr while working with the other software - but changing metadata in both applications for the same files at the same time will cause issues. If the automatic write-back is enabled in the Lr catalog it will flush out metadata whenever it sees fit, e.g. right after importing files. This will cause a re-read in IMatch and, depending on your IMatch configuration, also a write-back. I have figured out that using both programs at the same time it is best to disable the automatic write-back in Lr and make it write-back only at specific times.

I also find that, with all the new features related to face annotations (see I can annotate faces in IMatch super-fast, even without any kind of face recognition feature. It takes me the same time to add a face tag to an annotation created by IMatch it takes me to acknowledge the face tag suggested by Lr  ;D
-- Mario
IMatch Developer
Forum Administrator  -  Contact & Support - Follow me on 𝕏 - Like on Facebook



I am re-looking at iMatch and was a long term prior user, who switched to LR.

Would you clarify to what degree the face match features of LR and iMatch are compatible?  More specifically:

1. if a image is face-tagged in LR and then the face-tags are edited in IM, does LR update correctly (assuming metadata read/write is done correctly by user)?
2. for images face-tagged in IM 1st and not in LR, do those tags carry over correctly to LR?  Here I am asking if the actual face region gets the face-tag as if done in LR, not just if the person name keyword is updated.   
3. Overall would you say face-tagging is fully or very compatible between the two programs, or would that be overstating the compatibility.




I recently tagged a lot of scanned family pictures and slides. I usually started in LR and did most of the face tagging there, processed the images in DXO and imported them to Imatch. Mostly for testing I went back and forth between Imatch and LR and did not see a problem either. With some recent changes in Imatch the tagging got much easier and I don't use LR as much. Some others posted that LR doesn't update the tags, and that removing and adding the pictures back to the catalog resolves the issue. But that sounds like a LR issue and I never saw this issue. In my opinion the face tagging is fully compatible between the 2 programs, but LR has some features you will not find in Imatch at the moment.
If you work with keywords, you should be aware that the keywords are not linked to the annotation. If you change a keyword in Imatch, the corresponding annotation is not updated. I don't know how that works in LR, I don't use keywords.

If you just concerned about face tagging, currently LR may be a tick ahead of Imatch. If want to manage thousands of pictures within a real database and have control over metadata, Imatch is the way to go. I suggest you create a small test database with a set of  test pictures and try it out. Also, search for face tagging in the forum, there are some recent threads about face tagging with more information.



1. Yes. IMatch updates the XMP regions in the metadata when you change the face tag.

Lr has to be forced to reload the changed metadata via the right-click context menu ("Load Metadata"), it does not automatically pick up changes done in other applications.

2. Yes. Same principle as 1.

At least if you let IMatch also create keywords from your tags.
There have been reports about Lr versions which failed to import XMP regions when there was no keyword matching the tag.
Adobe should have fixed this bug by now. I guess.

I have to say that Adobe's sometimes works in mysterious ways, and that their software is not primarily designed to be cooperative - except to other Adobe products. IMatch goes a long way to be open and cooperative with all other software. This is not necessarily true for all applications out there, sadly.

Quote3. Overall would you say face-tagging is fully or very compatible between the two programs, or would that be overstating the compatibility.

Can't say, really. IMatch sticks to the Metadata Working Group (MWG) standards and specifications for face regions. This should ensure optimal interoperability with all other applications. At least the face regions are stored in a standardized and documented way, inside the easily accessible XMP format.


Microsoft, who is part of the MWG, decided not to stick to their own standard and instead uses a proprietary way to store face data produced by their applications. The face data is stored inside a proprietary Microsoft namespace in the XMP record.

It is conceivable that Adobe may go the same way, when they see a business purpose for it. Sad but true. May happen, or not.

The big companies all play a game to get your data into their hands ("free" cloud storage, proprietary formats, encrypting data to counter the "fair use" law, etc.). I think the general idea is to make it very easy/cheap for users to move their data into the data centers of the big companies and then make it very hard/expensive  to get it out again.
-- Mario
IMatch Developer
Forum Administrator  -  Contact & Support - Follow me on 𝕏 - Like on Facebook



Regarding interaction with LR and IMatch:

For a long time I used LR for developing my RAW files with limited use of the DAM side except to find images (that were tagged in IMatch) and perhaps change labels or ratings for Workflow tracking between the two apps.

I do not automatically write XMP data in LR (but I do in IMatch).  What i've found is that if I do something in IMatch while potentially having unwritten XMP data in LR, including edits which get saved to XMP, that LR will put a little "!" in the thumbnail for the image indicating that there is potentially conflicting XMP data. 

This is usually the conditions where I have problems as LR does not necessarily read in new XMP data unless you manually resync (rescan) your library.  I suspect Adobe doesn't believe there should be metadata changes outside its catalog, so it doesn't scan again... Although, I maybe a bit naive as there may be a setting for LR to continually scan for changes. 

I'm currently transitioning out of LR to go to DxO for processing my RAW files.  It'll avoid LR's catalog completely.  I've never liked that I have to go find images in LR after putting all the effort in IMatch.  LR does not make browsing my metadata that easy.