So what is Riya?
Riya is a tool that searches images and recognizes people and objects. By identifying all the people in your picture set, you can search on a person's name to find all the pictures of that person.
How does Riya do this?
First: You have to upload a huge number of photos (1000+) using an uploader application that is installed locally. Riya recommends uploading all your pictures over night or when you are away from your machine (whatever that is for you).
Local uploader your say, but this is Web 2.0? That's what I thought but the local uploader is required to use your PC cycles to do some pre-processing on the images. And hey, a local win32 app, its kind of kitschy in a 1998 sort of way. This is definitely the most awkward part of the product but it would seem technically unavoidable at this stage.
Second: As pictures get uploaded, Riya starts to go through them to identify faces. Then it asks you, Ms. User to label each face it has found. Once you give it a bunch of names to the faces, it starts to go into an auto-recognize cycle. It looks for those faces within other faces.
Three:.Riya asks you to confirm that it auto-recognized those faces correctly. It is really a 1,2,3 step process. When you keep repeating it, Riya keeps getting better and better at recognizing people correctly. Essentially, Riya starts to learn how to recognize the faces from your set.
How hard is it to recognize faces? Its hard for a computer, easy for a human. Have you ever sat with a parent, grandparent or sibling looking at old photos? I've seen my dad look at a picture from 1912 point at a boy in short pants and say 'That's my uncle Eddy'. My dad wouldn't have known his Uncle Eddy in short pants so how did he recognize him? The angle of the head, the area around his eyes, perhaps the smile made him see his adult uncle in the child. For a computer to do that type of recognition means converting these types of patterns into algorithms and then calculating probabilities of pattern matches.
Though Riya involves a 3 step process, the UI is fantastic and is the product's saving grace. Users barely tolerate 2 step processes. Riya uses some nice UI techniques to keep users on track.
The face recognition is good but Riya needs a large number of photos before it can reliably recognize a face. So at first, it seems like 'what's the point?' because it can barely recognize people and you have to keep telling it who they are over and over again. But as it starts to get better, it becomes more fun.
I like when Riya matches pictures of me to my sister or my niece to her brother, this makes it seem like the application is getting smarter. The T. and I had a laugh when Riya matched the pictures of two friends who we always joke are so alike in personality (turns out they look the same, algorithmically speaking).
A great thing about the tool is that it looks for all the faces in your photos. But a frustrating thing of using the tool is that it goes through pictures and looks for ALL the faces. Think of all the faces of people you don't know that you are forced to include in a wedding shot.
There were alot of people matched to my friends and family for which I thought, 'Who is that and why do I have a picture of them?' Riya has a 'Don't Know' option but I think they should add a 'Don't Care'. I would love it if eventually Riya could get smart enough to determine that a face that appears once doesn't have to be identified as quickly as one that appears many times.
And why does it keep matching me to my friend Alex? He's a dude, for one thing. And I keep telling it, 'Bad Riya. Bad match'. But it keeps bringing it back.
The nice touches of Riya outweigh the small problems. For example, Riya included pictures of my niece playing hockey in her name search. Not because it could match any faces (hockey helmets!) but because her name is in the album name and it included the album name in its Search.
Another cool feature of Riya is that it can be taught to recognize objects. In one picture of our living room, it identified our tv and our sony player. This was the first thing that impressed the T. When I first talked about Riya, the T. was less than intrigued because he has seen one too many projects that pursue an academically difficult problem that has little practical application. But to see it identify a TV won him over a bit. Why? Because its useful and I could point out to him that the collective had taught Riya to recognize TVs and now she was good at it. And she will only improve over time.
I have personally taken on the mission to teach Riya how to identify the CN Tower because I have alot of pictures of it. And I think the idea that I can teach Riya to identify this landmark is pretty cool.
So how did the collective teach Riya? Because on upload, all pictures are public! So it compares my TV picture against all the other TV pictures. This surprised me as I thought it should be private by default. I don't put pictures of my friends and family public without their permission. I suspect that the majority of people would share this view with me. But the lead users of Riya (who routinely put all pics up on Flickr) probably don't care about this. But that default setting is something Team Riya should probably think about more if they are going for wider adoption.
So I'm hooked on Riya's wash, rinse, repeat cycle. There are pictures I haven't looked at in years. I can see how badly my photo albums were organized and now with Riya I found my niece's hockey pictures. My only question is ... Now what?
This leads into a question about what will Riya's business model be after it leaves Beta? I will put my thoughts into another post.