Picasa has Face Detection on the Desktop (for real this time)

A little over a year ago I rushed to post about the latest release of Picasa, Google's free photo browser/manager/editor and their introduction of a Face Recognition feature. It turned out that my post was a bit premature in that they had actually only released this feature on the web version of the tool... well, today that has all changed - the 3.5 desktop release includes the mug finder!

Picasa Supports DNG - but can we use the face tags as keywords?
Now, this is all well and good for Picasa users, but I use Lightroom to manage my images. What I had written previously still applies - I'd love to find a way to work this into my workflow! Fortunately, the Picasa software supports popular RAW formats - including the Adobe DNG format. This means that you can turn Picasa loose on your image library and let it identify people in your photographs.

But - Picasa doesn't seem to save the name tags in the metadata of the files... this is a bummer, since it means we can use the tagging in other applications. Ideally, you would want the head tags to be added as keywords in some standard format - say person:first_last or something like that. I'm not too familiar with the ins and outs of Picasa, so this may be possible - more experimentation is necessary (I've not seen it in any obvious place so far). Another possibility would be to write a quick script that will automate the process of reading the head tags from Picasa and then updating the image EXIF with the keywords...

If anyone does, or knows of a way to do something like this please post in the comments!

About the Face Recognition Feature Performance
After downloading the software I let Picasa import the DNGs of a large wedding I'm currently working on... well over a thousand photos containing tons of faces and totaling 30GB or so. First thing to note: this is going to take a very... long... time. That's fine by me - I'd prefer that Picasa take the time to really look hard at the photos and figure out who's in them. Even if it works through the night, it'll certainly finish before I could - if it wasn't such a monumental task as to prevent my even attempting it in the first place!

The way the process works is pretty straightforward: Picasa scans all images and looks for FACES. If it knows that face, it'll take a shot at automatically tagging it. Otherwise, it will add the face to the "Album of Unnamed People" - a potter's field where you can go to assign names to faces. Clicking on a face will allow you to assign a name; from there Picasa will take this hint and apply it to the rest of the faces it has seen. The software appears to "learn" as it goes, so once you start giving it a few hints it'll get better and better at identifying the folks it comes across - in spite of lighting, pose or even focus (not that there are many of those... er... well, perhaps a few).

So like I said - this will take some time. Picasa has been chugging away in the background burning through CPU cycles (2.8GHz core 2 duo w/ 6MB cache & 4GB ram)... it is only 10% through analyzing the 30GB of images in the hour or so it has been running. The results look promising thus far & I'm sure this will be a great tool once we figure out how to extract the information from Picasa and embed it into the metadata.

Congrats to the Google Picasa team on a job well done & thanks for this great new feature!