Dear iPhoto: Faces rocks, but I am not a bicycle

David Braue
11 February, 2009
View more articles fromthe author
AAA
Blogs

I have previously mentioned my enthusiasm for the new features in iLife ’09, but only – finally – got the chance to play with the bundle this weekend. Given my ongoing whinging about the mammoth task of tagging my huge photo library, I thought I would share my thoughts on iLife ’09′s much ballyhooed Faces feature.

It is, in a word, absolutely bloody amazing. Oh, that’s three words. Sorry.

Anyways. I installed iPhoto ’09, pointed it at my photo collection – which thanks to some mass importing has swelled to some 28,000 photos occupying over 60GB of disk space – and waited. And waited. And waited. Faces, you see, does an initial scan of your entire photo library to pick out every single face it can find. It doesn’t tell you how many it finds, but I estimate around 50,000 faces – and it took about 6 hours to complete.

But that’s a one-off. Once it was done, I loaded up a photo, clicked the Names button, and there was Faces – asking me who it was, just like in the screenshots. I typed in a name, clicked on the arrow, and within 20 seconds it had pulled up a list of other pictures that might feature the same person.

And it was totally, completely right. Mostly.

Scrolling down, I saw a few mismatches – 2 out of 55 images on the first page, 8 out of 55 on the second, 18 out of 55 on the third, and 23 out of 55 on the fourth. I then clicked ‘Done’, even though it had offered more photos to tag, and in so doing forced Faces to update its faces database and pull out a new list of potential matches. These proved to be a bit closer in accuracy, although accuracy decreased quickly as I scrolled down the list of well over 500 possible matches.

You learn quickly how to get efficient with Faces – I will share a few tips in a moment – but the results speak for themselves. Starting from zero, I had tagged over 1210 photos of my first person within 20 minutes, and after a few hours have tagged seven different people and over 4000 faces.

There’s still a long way to go, but I am no longer filled with dread at the thought of doing it because I can check a huge number of  images at one time. Compare that with the alternative – using keywords and spending hours scanning the picture, typing keyword shortcuts for each person involved, and then pressing the right arrow to move to the next picture. Because Faces only shows me faces – and not the context they are in – I don’t get distracted or start marvelling at a picture, or have the urge to design a book, or anything. It is just pure, unadulterated tagging – a yes or no for each picture, and nothing else.

Many reviews of Faces have made light of the feature, with our own Jim Dalrymple recently amused at its insistence that he is Steve Jobs. Such lighthearted comparisons are easy to make – after all, in my adventures Faces picked one face out of an embroidered pillow – but the fact is that Faces can only work with what it’s given. And if your library is filled with pictures of Steve Jobs, well, that’s what it is given.

The technology Apple uses will no doubt be tweaked in future versions, but for my money it is the most startlingly amazing thing you will have seen done in software for some time.

iPhoto correctly picked faces of people that were looking at the camera; looking down; rotated 90, 180 or any other number of degrees; underexposed; overexposed; blurry; grainy; and combinations of these. It picked people even if they were wearing glasses, sunglasses, hats, scarves, helmets, wigs, and even when their faces were partially obscured by someone’s shoulder, their hair, and so on. It even picks faces out of photos that might be on a table in the background of your real photo. It even made a correct match in a picture where no part of the face below the nose was showing.

Indeed, the secret seems to lie somewhere in the shape of the nose and its relationship to the eyes: in some pictures, little more than the nose sticking out would still generate a match. This would also explain why iPhoto can’t recognise pets or stuffed animals: their wet, black snouts simply do not correspond to human features (note to Apple: pet lovers everywhere will love you if you can work the same magic with Fido or Mittens).

Much has been made about whether Faces can track children as they age, and let me say that it can: a number of under-10 children in the extended family were identified correctly – using seed photos taken within the past few weeks – in photos taken as many as seven years ago. Age does not seem to be an issue at all.

Now, before I outdo myself with praise, let me say that Faces is not perfect. After my first stunningly successful run, I started with another person who was far less represented in my photo library, and Faces floundered a bit. By picking out matches and clicking Done – and repeating this several times – accuracy seemed to gradually improve.
When I set it to recognising my own face, accuracy was quite good – only 15 mistakes (including, incredibly, a bicycle) out of 236 photos. And it took me less than a minute to find and fix them all.

I mentioned I had a few tips to speed this process – and, if you have anywhere near as many photos as me, you’ll want to speed it as much as possible. Here you go:

  • When confirming which faces are correct, you can drag a marquee around a group of photos to confirm they are the person in question. This is not just a nice-to-have.
  • When you’re choosing the seed photo for each person, pick one that shows a clear frontal view without any part of the face obscured; this seemed to provide strong matches.
  • Start by tagging the person you photograph the most; this will not only help you get your library under control more quickly, but it will eliminate many possible but incorrect face matches when you’re tagging other people.
  • Hold down Ctrl and the tagging option changes to ‘Reject’; each photo you click will now be tagged as not-a-match. This saves you from double-clicking as described in some online reviews. You can also hold down Ctrl while dragging a marquee, if you see a large group of photos that aren’t matches.
  • Instead of reviewing one image at a time, select the entire screenful of faces as either a yes-match (green) or no-match (red) using the marquee tool; then, quickly scan each row and click once on the exceptions. In other words, if it looks like there are more matches than non-matches, designate all as matches then run through the images and only click once on each one that’s not a match.
  • Scan the photos with your mouse so you can just click when you need to and move on. Getting out of your zone will only slow you down.
  • If you change the size of the thumbnails, you can review loads of photos at once; their diminutive size isn’t a problem because Faces only shows you the face in question. On my 24″ iMac screen, at 1920×1200 resolution, I could squeeze up to 15 faces across by 6 rows for a total of 90 images at once. Needless to say, this speeds up the tagging process considerably.
  • Once you’re done tagging a person – when Faces can’t find any more possible matches – I suggest you do a quick select all (Option-A), turn on keywords (Option-K), then tag all the photos in the Faces view with a real keyword matching that person’s name. This will make searching and sorting extremely easy in the future.

If you happen to work for Apple or know someone who does, here are a few small requests for the next version of Faces:

  • Allow specification of hair colour to give weighted preference to certain matches: if the person I’m tagging is a brunette, it would be nice to be able to exclude all blondes if I want to.
  • If two or more people tend to be photographed together – sisters, partners or families, for example – it would be great if Faces could be informed of these relationships and weight results accordingly.
  • A keyword search on the suggested matches could be useful for quickly tagging: for example, searching for ‘christening’ would pull up all the photos from a baby’s Christening, narrowing the decision set significantly. Allowing filtering by date would allow exclusion of pictures taken before a child was born, eliminating many false matches.
  • It would be nice to have toggles to search for people with glasses, or closed eyes, or wigs, or smiles, or redeye, or faces covered, or so on; surely Faces’ recognition could easily discern such features, and it would be invaluable in sorting and selecting photos.
  • When reviewing possible matches, it would be great to be able to go through the photos and only tag those that are matches – then use an option to tag the rest as non-matches.
  • Why stop with faces? It would be great to be able to draw a box around a feature in a picture – a building, for example, or a stuffed animal – then find all photos that appear to have that same feature in them. This sort of photo matching is already available in some photo managers, and would seem to be a small step for iPhoto.

Obviously, your mileage will vary. But after just a few hours using Faces, I’m pleased to report that it has lived up to expectations – and raised the glimmer of hope that I might possibly, one day, get all these photos tagged.

Leave a Comment

Please keep your comments friendly on the topic.

Contact us