Quick review of iPhoto 09 Places and Faces

I received my copy of Apple's new iLife 09 a couple of days ago, and being a keen photographer have spent quite a bit of time trying out the new features of iPhoto, especially Places and Faces. This is just a quick initial review - I may do a more detailed one in the next few days. Briefly, I very much like Places, with the exception of a few minor niggles, but Faces, despite an undeniable coolness when it correctly identifies someone, was overall a disappointment.

Places does a nice job of mapping geotagged photos. What I especially liked was the way that it clusters groups of photos that are close into a single pin on the map as you zoom out, and it breaks them out as you zoom it. Some other systems out there do this too, but most that I have tried have some issues (for example only displaying a limited number of pictures on a single map). I haven't done serious scalability testing so far, but have about 1700 geotagged photos on my world map so far and performance is good (these are a mixture of precisely geotagged photos with a latitude-longitude from a GPS stored in the photos' EXIF metadata, and others where I have just specified the city for a group of photos). It is easy to specify a location for a single photo or a group of photos, using Google search for geocoding so you can find the usual assortment of landmarks, businesses, etc as well as addresses. The system also generates a nice hierarchy of place names, so for example I can ask for all the photos in Stanley Park, Vancouver, British Columbia or Canada. There are some situations where the reverse geocoding doesn't find a name at some level of the hierarchy, and it uses a label of "Other", but overall this works well. Overall I think it's the nicest implementation of a system for viewing and managing geotagged photos that I have seen so far.

I really wanted to like Faces - the demo certainly looked really cool. I take a lot of photos and manage them in Aperture rather than iPhoto (I currently have 108,000 images in Aperture), and whenever I have time I tag the photos, including names of people in them. So good face recognition has the potential to be huge time saver. It's really cool when it works, and it does work well a good amount of the time, but overall it didn't work correctly sufficiently often that I'm not sure it is as fast as manually tagging names in Aperture. I will probably try to do a more systematic test of this impression over the weekend. There are two ways it can not work - it can find a face in the photo but not know who it is, or it can fail to find a face in the picture at all. The second case happened more often than I expected - it did very poorly with people wearing hats, large sunglasses, or with faces at an angle. Adding a face manually is a lot slower than doing a manual tag in Aperture (which can be a single key press for "favorite" tags, or a single drag and drop otherwise) - you have to click a button, drag to move a box, and drag again to resize a box, click a button, and type two or three characters of a name. There's also a confirmation process where you click to confirm that faces are who the system thinks they are, and (as far as I can tell) this has no ability to confirm multiple pictures at once (for example clicking on the first of a sequence and then shift-clicking on the last, as you normally do to select multiple items) - you have to click on each individual picture. The system could also be a lot cleverer about handling pictures that are taken in a sequence. It is very common with digital photography to take several photos of a group of people in a burst - and the date and time are stored so this situation is easy to identify. If you manually identify people in the first shot of a sequence, this should really carry through to other photos in the sequence, but it doesn't.

Also, as with iPhoto in general, Faces does not seem geared to handling large number of photos. I have 50+ people in my "Faces" already, a lot of whom just appear in a few photos, and there's no easy way to filter these down or organize them into groups etc, so it will be pretty unwieldy as I load more in there.

So overall, I think face recognition is a really exciting technology and it will be hugely useful in the future, but I don't think it's quite there yet based on my experience so far. In addition to improvements in the core recognition technology, there's a lot that Apple could do to make the workflow more efficient - it definitely has the feel of a "1.0" release (which is a little disappointing coming from Apple, with their strong focus on usability). I hope to see the face recognition capability come to Aperture soon though, hopefully with some improvements in the workflow for large volumes of pictures.