How can I force Picasa to rescan faces?

It is possible to jolt the system to at least present more, different suggestions to the user. One can do this by adjusting the ‘suggestion threshold’ and ‘cluster threshold’ in the preferences or tools area of the application.

London 2010 Basketball arena

Picasa is Google’s free photo management application.  It appears to be an experimental project rather than a flagship product, but its extremely useful and versatile.  In particular, it allows management of photos without making a copy of each photograph inside the application.  This was the feature that prompted me to move my photo management into Picasa from Apple’s iPhoto (now discontinued) and why I have not moved on to its new ‘Photos’ offering.

Like those other programmes, Picasa has a powerful facial recognition feature.  Set it loose on your photos and it will recognise faces within them.  The software if powerful enough to identify blurred, grainy faces as well as in-focus portraits.  If you give a face a name, it will identify other similar faces and suggest that they are the same person.  I began tagging all my photographs like this.  As an obsessive archivist I have just over 22,000 unique photos in my library (I know they’re unique because I have removed duplicates using PhotoSweeper from Overmacs).  The Is There A Face In This Image? functionality worked as it should throught—even on ridiculous many-faceted images like the one featured above.  The programme took about 5 hours to scan every photograph, but I am confident it identified all the faces in all the photographs in my library.  There were over 30,000!

The Whose Face Is This? functionality initially seemed to work extremely well.  The software was spookily efficient at correctly identifying images of the same person.  Picasa offers a handy ‘confirm all’ button, so when it suggests that a further dozen or score of faces might all be of the same person, one can tag many faces at once. It did not take long to tag a few thousand photos.

And that’s when the software seemed to fall over.  Scanning the as-yet unidentified faces, I can see many faces of family members who have already been tagged, and who the software should already ‘know’.  Despite this, it no longer suggests names for those faces.  On reflection, this is unsurprising, because for the recognition to take place, the programme needs to compare each of (say) thirty thousand faces with each of the other thirty thousand faces.  That is nine hundred million comparisons!

Instead, Picasa presents only a few faces for tagging.  When these are labelled, it seems to present more in an unpredictable and ad-hoc manner.  It no longer seems to learn.

Searching online, I have found no reliable suggestions for kick-starting the scanning process, so that Picasa applies what it has learnt about tagged faces to those currently without a name.

Turning the features off then on a gain is a no-no.  That just forces the system to start from scratch!  Previously tagged images have to be tagged again.

However, it is possible to jolt the system to at least present more, different suggestions to the user.  One can do this by adjusting the ‘suggestion threshold’ and ‘cluster threshold’ in the preferences or tools area of the application.

Screen Shot 2015-09-17 at 16.16.11This does not fix the problem perfectly but does prompt the system to present more suggestions to the user, allowing for quicker tagging of the backlog.

Unfortunately, there does not seem to be a particularly vibrant community of people developing fixes or plugins for Picasa.  With an entirely open-source project, one would have expected there to be some plugin or patch that allows the facial recognition features to work better at high volume.  This does not seem to be forthcoming.  One can hardly complain, given that the entire thing is free to download.

Nevertheless, if anyone with more experience with Picasa can offer suggestions on how to unleash the potential of its clever facial recognition features when there are tens of thousands of faces identified in the library, I would be very grateful.

Leave a Reply

Your email address will not be published. Required fields are marked *