Authentically Human! Not Written by AI!
All Content Copyright © Michael Chesley Johnson AIS PSA MPAC

Wednesday, June 27, 2012

Cataloguing Digital Reference Photos


"Luminous Fog" 6x8, oil
If I were to apply labels to this, I might use "fog," "tree," "water," "cliff."  But what about applying
labels that would indicate mood, color scheme or key?

If you're like me, you're very bad about organizing your digital reference photos.  I take a few hundred pictures at a time, dump them to my hard drive and then forget about them.  I put them into folders that are stamped with the date on which I took the photos.  Back years ago, when I didn't have so many photos, it was pretty easy to remember approximately when I made a particular field trip and took a photo of a particular subject.  That great photo of those herring smokehouses with the tide out on Grand Manan? Yeah, that's right, I think it was the middle of May, 2005.  I have a great visual memory.

I could find that photo in 2006 or 2007.  But seven years later, my own memory banks are too full to remember when I made that trip to Grand Manan.  I have to go through my dated folders, surfing through tiny thumbnails.

If I'd been smart about it, I would have started labelling photos from the start.  You could always label photos with Adobe Photoshop, and recently, Google's Picasa added this capability.   (I use Picasa to search my computer and put images in albums.)  But it's a Herculean task - maybe more like a Sysiphean one - to do this retroactively.  I don't have the time to go back and apply labels.

And of course, I still don't label my new photos.

Google is working on software that will, among other tasks, label photos for you automatically.  You may know that Picasa can already recognize if there is a face in an image and label it as such, but this is light-years beyond this.  Once the software is trained, it should be able to find that photo of smokehouses for me.  But it's going to require more computing power than I have in my desktop - or in yours.   The software currently is being trained on a supercomputer made up of 1600 CPUs.

Here's a news article about this developing neural network.  It has learned to recognize a cat when it sees one, without any human coaching.