Skip to main content
Participant
September 25, 2023
Question

Facial recognition now detecting everything

  • September 25, 2023
  • 4 replies
  • 3813 views

Ive got a catalogue of about 30,000 mixed media files all indexed and tagged with facial recognition. Ive been trying to clear up all my unmamed people and discovered that Lightroom has taken it upon itself to also identify about 90% of the surface of about 4000 pictures that all have previously identified faces, for example in a mountain scene I would have the people in the forground identified and then it tries encapsulates everything by trying to recognise the whole mountain range!!.

So the database is stuffed because its now peppered with these erroneous recognition areas and Ive got a horrible feeling that I'm going to have to identify and delete every one of the faulty zones.

 

I've tried using Adobe support, but honestly, its embarasing in the extreme and its wasted 3 hours of my life achieving nothing and being none the wiser as they work through their TS script for the Adobe Illiteratae

 

Dont know when the corruption krept in so going back through old backups would be possible but very time consuming and is the "Hail Mary" option so grateful for any assistance and also if anyone has encountered this , how to prevent it in the future as Adobe have not the slightest interest

 

I've attached an example of the incorect facial zoning

This topic has been closed for replies.

4 replies

Participant
July 6, 2024

I can tell you cheap software can cause alot of problems with your cameras my brother had a cheap system from Walmart first mistake. But anyway had it pointed at backdoor inside a fencd in yard with a locked gate but about twice a week it was showing somebody standing there that he didn't recognize and nobody climb the fence and went to the back door anyway so only thing we could come up with is that it was picking people up from the neighbors backyard heard other story's like this with people using similar systems you get what you pay for I guess

johnrellis
Legend
September 25, 2023

Something is seriously wrong with your LR installation and/or catalog. While LR Classic's face recognition has stagnated in recent years, its identification of face rectangles is generally good and never as bad as your example. I imported your screenshot into my LR, and it doesn't identify that flower as a face.

 

As a next step, select several of these bad photos and do Library > Find Faces Again, leaving both options unchecked. Do you get the same bad face rectangles?

Participant
September 27, 2023

Hi John ........ OK, so I ran your test on several different types of photos (including some Water Buffalos! ).  The old recognition data was cleared out and replaced with new.  The new data only contained faces of people (Phew , no water Buffaloes! ) so the FR looks like its now correctly identifying faces again.  After seven hours with support I managed to actually speak to someone who begrudginly agreed to escalate the problem.  They have all of the requested debug info so I'm waiting on them to come back to me, but to be honest I'm not holding my breath.  I guess Im going to put some time into figuring out how to purge all of the erronous entries from the database do you have any info on database structures, structured queies and any dbase sanatising tools for LR?

johnrellis
Legend
September 27, 2023

Based on years of observing the LR team, I would be surprised if you got anything useful from support in the end. They're more miss than hit.

 

You could use the Any Filter plugin to find all very large face rectangles. E.g to find all face rectangles that are least 50% of the total photo area:

 

The most expedient solution might be to take an hour or two and delete those bogus rectangles. 

 

Another approach would be to go to People view, select some or all the thumbnails that appear under Unnamed, and type Delete, which deletes the face rectangles.  You'll be able to see thumbnails of the rectangles, letting you quickly decide if you want to keep some of them..

 

The catalog database isn't documented. It uses Sqlite (the widely used, very robust open-source database library), and there are numerous open-source tools for examining Sqlite databases: sqlite3 (the command-line utility), DB Browser for Sqlite, and many others.   People experienced with relational databases are typically able to figure out the structure of the catalog for what they want to do in an hour or two.

 

But whichever method(s) you try, be sure to backup your catalog first (and make sure the backup actually works)!

 

 

 

Participant
September 25, 2023

just created a new catalogue rescanned all images for faces and look what facial recognition thinks is a face

Community Expert
September 25, 2023

The facial recognition in Classic has been quite neglected for years. It's very slow and buggy. I think they gave up on fixing it because Lightroom Cloudy does AI face recgnition in the cloud now which works much better but is quite unlikely (i.e. hell freezes over) to ever be ported to Classic. That said, if you have lots of unknown faces with question marks, just ignore them. As long as you don't actually acknowledge them, they won't show up in the faces it knows about. I have 100's of these. Most I have no clue how it could ever think there was a face there.

Participant
September 25, 2023

Jao .... I could switch but I've never really though about the cloud version is it a hybrid i.e all storage local and processing shared between local hosts and cloud?  and what happens when no connection ?

Community Expert
September 25, 2023

I wasn't suggesting you switch! Just that Adobe seems to not be intersted in fixing this in Classic so it is probably we'll have to live with in Classic. Cloudy is based on the idea that all images are in the cloud. It uploads all your images and dynamically downloads copies of images when needed. Processing is done on your local machine but things like AI based object and face recognition are done on the cloud. All this means indeed that if you don't have connectivity things break a bit. Cloudy is much less full featured than Classic in other ways (it can't print and many more things are missing) so I generally do not recommend it except if it is very clear you don't need any of the missing features or the cloud synchronization of devices is so attractive that it doesn't matter to you that those are missing.