Why I’m Disappointed in Today’s News on Riya
[I am the Evangelist and CEO of Zooomr]
Techcrunch About a year ago when Riya first launched with a big splash as a new Web 2.0 company with super exciting facial recognition software I was excited. I was super excited. Like a lot of people I uploaded a bunch of my photos to their site. I really liked their bulk uploader, found the site’s navigation nice, and then proceeded to spend hours and hours and hours “training” Riya on the faces that I uploaded.
The promise in Riya’s technology for me was that by training Riya on some of my photos that Riya could then find other photos of mine of the same people and auto tag them for me. Since I take a *ton* of photos, this technology held particular promise. More than this though the technology that held the most promise to me was that of my training Riya on me, my family and friends, and then being able to use it to search the internet and find photos of me, my family and friends that I did not know existed.
I was particularly excited about the idea of combining Riya’s facial recognition technology with social networking at the time and sent a bunch of emails to Tara Hunt back when she worked there about how cool Riya would be as a platform for a social network. Given the early enthusiasm for the site and the massive uploads they were getting, building a social network around Riya seemed like a no brainer to me.
I then was *really* intrigued when I heard that they were in play with Google and this made me think that their technology really might work.
Unfortunately their technology did not work very well. Even after training Riya on multiple instances of faces it did a poor job at finding other photos of those individuals within my own library, let alone out on the web.
Where some people get excited about sports or celebrities or cars or politics, I on the other hand get excited about image search. I get really excited about image search. And so I figured that it would just take time for Riya to continue to refine their facial recognition software and get it to a point where it was good enough to use it to effectively use it for image search.
And then today we get the news that instead of focusing on people search Riya is turning themselves into some kind of a shopping site — allowing you to find photos of black purses with suede handles or something like that and then buy them. They are calling it Like.com and it’s up right now and you can try it out.
While I can appreciate that Riya had to come up with some kind of a business model to justify all the money that has been invested in the company, I still find this very personally disappointing. I find it disappointing because 1. It means that facial recognition search technology likely cannot be automated in the short run and 2. I just have no interest in finding photos of slip on red loafers with black tassels or Paris Hilton’s latest watch to buy. According to Shah, “Like searches merchants sites for Jewelry, Handbags, Shoes, Watches. 200 merchants 2M SKUs.”
I know that there is probably just a huge market for people who want to buy things this way, I’m just not it. I have no interest in searching Riya’s two million SKUs. I want to search “Las Vegas” for the best shots when I go and visit. And I want to ego search “Thomas Hawk” to see what images of me are up and out there on the internet. I want to search “Bruce Livingstone” iStockphotos’s CEO because I met him two weeks ago and thought he had interesting tattoos (do you know he actually has an iStockphoto tattoo?) . I want to search for beautiful shots of rain, and Brazil and Antartica, and Charles Bukowski and San Francisco’s favorite eccentric Frank Chu, and “public art” and Banksy images, etc. etc.
The problem is that I have no interest in doing jewelery searches.
Scoble also says that Riya just couldn’t get facial recognition software good enough to make it work. “Why not keep working on face detection? Because they learned through user testing that they’d never be able to make it good enough. They found that by focusing on visual image searches they can get a much more satisfied user base.”
And maybe this is the reason why a rumored $40 million buyout from Google never happened.
The question of how image search ought to best be done is an important one. Right now Google and Yahoo, the two primary players pretty much suck at it. Both Google and Yahoo are trying to improve though. Google with their Image Labeler game (that I’m not particularly fond of) and Yahoo with Flickr (which in my opinion is the best image search engine out today). Getty has some interesting search tools in the works which would hold promise to the extent that they have access to high quality image libraries. Ask and Microsoft are also trying to get into the game, but neither in my opinion are better than Yahoo or Google and both here too pale in comparison to Flickr.
So it would appear that social based image search still represents the best approach. This is what we are working on at Zooomr. In terms of people search we are using an approach called people tags which have you actually create a notation around an individual where you can people tag them. These images are then grouped with that individual on their profile and you can see photos of them. It does rely on the social network and manually tagging though vs. the promise that Riya held with facial recognition. Here is an example of photos that have been people tagged with Thomas Hawk so far.
Some of these are self portraits taken by me, others are photos of me taken by my friends and contacts. Over time this collection of images of me should grow and grow. I can also use SmartSets to filter photos further. If for instance I only want to see photos that Kristopher Tate and I are in together I can create a SmartSet of those images based on the people tags — which you can see here.
Much of this technology will be of more use to individuals as ways to organize photos of their family and friends than for general web search, but it will certainly have general web search application as well as cluster’s of photos of well known people certainly will appear.
Image search is going to get more and more important in the future. Particularly local image search that is tied to local businesses. Finding photos of a particular restaurant or night club that you are interested in for instance. Or photos of a park that you want to visit. We are working on these technologies too.
Maybe I’m just naive and shopping image search is really the most lucrative sweet spot of the image search game. But to me it’s not as rewarding or as personally satisfying as what I would have hoped would have come out of Riya.
Update: more from Dan Farber at CNET here.