Slashdot Mirror
Riya Eases Pain of Digital Image Management
Vitaly Friedman writes to tell us Wired is covering a new service that hopes to alleviate some of the woes of digital picture management using face-recognition technology. Riya, requires a bit of upfront training but thereafter it is able to identify and tag individuals in your pictures along with text recognition for street signs and the like. The service also plans to offer the ability to make your online photo albums public, private, or viewable by invitation.
It's almost as if they were exactly prepared for a slashdotting! Very clever people, indeed.
John
There should be a law against sites which require IE6. ... thing.
I knew it was a Win32 application, I was ready to try it with CEDEGA, but to register you need Internet Explorer 6. I really don't like this kind of
Most of my photos are upskirts of random strangers at the mall. This service is only useful if I'm going to be photographing the same shaved vaginas over and over again and, for that to happen, I'd have to be incredibly lucky. You try asking a hottie out when you have mirrors on your shoes and a camera with the "record" light on, aimed upward from knee-height.
Wait, that is information about him. Nevermind then.
public class null extends java applet { System.out.print ("Tabula Rasa"); }
Grab a shot from your office security camera of the guy that just came in for a job interview. Use Google Image search now with Riya and find all those photos of him passed out drunk in college his buddies posted to their Flickr account, or the nude images he posted to "that" site, or how he's doing on Hot or Not.
Wow. What dufus marked your post "offtopic"...!?
Personally, I thought MySpace was for college students. Or really, anyone young enough to still be intrigued by being part of a "clique".
Imagine what happens when lots of people in public places use their webcams (the kind in most cities or small shops) to record and upload every person that walks by. You'd eventually have a huge public database of automatically catalogued photographic timelines of every person. Where they went, when, who they were with... yikes.
I can't remember all the times I've sobered up after a weekend bender, only to download the camera and ask myself "Jesus, whose **** is that???"
This will make it so much easier to find all the right people when I have to make that awkward call from the clinic.
Here's a thought for an interesting/scary feature: The ability to search across all public albums for a particular person within a given timeframe, based upon the timestamp of the photograph. So you could say "Find pictures of Sue from Dec 2 to Dec 6". Now maybe Sue was supposed to be 'out of town', but you find her with someone else...
;)
This should be fun.
Be very, very careful what you put into that head, because you will never, ever get it out. -Thomas Cardinal Wolsey
When I say that the "Pain of Digital Image Management" is mostly RSI.
No folly is more costly than the folly of intolerant idealism. - Winston Churchill
Castera also wonders if Riya might be useful for paternity tests.
"I submitted some photos of a little boy and others of his father -- who is my best friend -- and Riya found the resemblance," said Castera, a French industrial designer. "It's very touching."
Trusting some unknown face recognition software to do "paternity tests" is a little out there.
Isn't that how paternity was determined back in old days? ("'E's the spitting image of the Duke 'e is.")
I don't think he's kidding.
Send email from the afterlife! Write your e-will at Dead Man's Switch.
Thats because its not the same thing. Facebook cannot be trained to recognize faces. You have to tag it for every picture manually. The only doofus is you.
The two are not really alike.
is Slashdot going to start promoting everyone's new photo sharing website? Like most, I clicked on this, tried to sign up, and "We only support IE6". (Me dragging Riya to Recycle Bin, Empty Recycle Bin)
------ Tim O'Brien
Google will buy this company. This would be great for intergration with their already pritty good photo software. I like the idea of being able to seach my harddrive for all photos of certian people from along time ago.
I know this probably won't interest Americans, but if this type of software can read street signs- can we also assume it will recognize traffic signs and house numbers?
A lot of countries don't have access to free information, so this technology if it worked would seriously harm some of the mapping data monopolies.
Information: "I want to be anthropomorphized"
Of course they're the same thing (essentially). One just adds the functionality of saying "if the other 100 pictures sharing these certain characteristics were labeled with these specific tags, then this one should be as well" rather than requiring the 101st photo to be manually tagged.
Geez, boob recognition software would matter more to the /. crowd. No one looks at faces here.
Thats because its not the same thing. Facebook cannot be trained to recognize faces. You have to tag it for every picture manually. The only doofus is you.
I'll bet in actual practice they are largely the same. I read this story and the "Begging for VC funding by grossly overstating capabilities" flag in my mind was waving at full mast. Face recognition is a technology in the infantile stage, and every demo that I've ever seen has used a database with just a couple of largely distinct faces: In reality with a large group of similar faces under less than ideal scenarios, it is close to useless. I know a lot of large organizations and security shops are being sold the face rec B.S., but face recognition in practice is close to useless.
Of course in this case I'm sure you're hoping that you "train" it with a set of just a few faces, and I'm sure then it could randomly guess and (if you have 4 faces) people would be amazed if it was right 25% of the time. Saying "oh it works but isn't really that accurate---but it's alpha!" is nonsense. They either have the technology, or they don't, and if it's marginally accurate (so you're spending most of your time correcting it) then it's truly worse than nothing.
Paternity tests...god...give me a break. This story has every hallmark of a scam. It sounds like the endless "super compression" stories.
Anusol eases the pain of digital rectal examinations by soothing haemorrhoids. Time for my pills.
Stick Men
At first i misread this for "digital rights management" i thought they had invented a DRM scheme that will only play music if the person who bought it is looking into the camra
It's calling Opera users as Firefox users! Don't ridicule us like that! ;-)
Seriously, I don't understand why they *block* browsers like that. So, it renders a bit wrong? Whoop-di-doo... Before they've started supporting these browsers, if it would get popular, someone has written e.g. a Greasemonkey script to fix it. I can imagine an Internet bank doing it for it not having passed a browser security test yet and they'd have such routines, but a photo search service? What terrible things could happen from a different web browser rendering the content slightly differently?
Beware: In C++, your friends can see your privates!
Anyone know offhand how they do the photo tagging? I use Picasa and do my tagging manually. I have a few hundred photos I just scanned and I don't want to manually tag them with names if there's an automatic way. Do any of these tagging programs use the same tagging format? I think Picasa uses data files in each directory and some kind of central database. I imagine each program does something proprietary so none of it carries over between programs, but I could be wrong and I haven't looked into it yet.
Perhaps someone more knowledgeable than me can say something about this? Thanks!
Stop the Slashdot Effect! Don't read the articles!
Actually, if it ever becomes a standalone product, there is substantial application potential for this technology in the domains of cult film and adult video. For instance, there are many cult film and adult video databases and communities that attempt to catalog and cross-reference performers. Tens of thousands of films and videos exhibit only threadbare production credits (or none at all), so huge numbers of performers are left all but anonymous.
For those who have attempted the incredibly daunting task of compiling filmographies by comparing stills captured from analog video sources, a technology like this--if even 30% reliable--would be surpassingly useful.
I don't really know what pica or riya do but it would not take much. Most image formats have room in the image itself for tags. You can see and set them with editors like gimp or hexedit. The name itself, if unique, is enough for most databases and the information could easily fit into the picture. All of it seems creepy and invasive to me.
I organize with directory structure hand and serve the results myself on my cable modem. I like my methods of organization better than any robot's and ALL of my pictures are shared. Only a few people are interested in them, so they get what they need despite the cable company's obnoxious upload speed crimps. The result will always work and I don't have to worry about a non free database format going away.
It might sound tedious but it's not. I've got a photo album with family, friends, house, do and user directories. Anytime I photograph an event I make a labled directory like 2005_09_12_uncle_visit and stick everything in it with an index made by konqueror or igal. The directory is moved to an appropriate directory on the cable box and symlinked to others, including a big chron directory with everything organized by year. Movie files are tedious, because my tools do not automatically index them. Finding things is as easy as looking in the right place or time. Occasionally, I tar the whole thing and move it to another drive to ward off dissaster. The whole indexing process takes about five minutes more than simply dumping the files would. Automating the process might be nice, but I'd never trust some non free software to do it.
These services touches on issues of free and non free versions of the web. In a free web, people have the tools to do everything themselves and it's easy. In the non free web, things are hidden, difficult and what you can't do is used as leverage to gain information that should be private. In a world with companies like Choice Point, my method gives my friends what they want but not the spammers. My friends can recognize the people in the pictures I have and are not very interested in pictures of people they don't know. They can find what they are interested in about as easily as I can but there are no files with people's names and other identifying information on them. In the non free world, people teach someone else's database the names, addresses and faces of all their friends and family for the ability to point to some a page full of adverts that won't be there in a few years.
Friends don't help friends install M$ junk.
If you had RTFA, you might have noticed that they key to it working is limiting data points. In other words, the population it works on is always small, just the people in the pictures you have in you picture collection. So it has a relatively small number of people to identify, but usually has lot of pictures of those people. It's the perfect kind of dataset for this kind of technology to work with. So maybe you should try to technology first, or at least try to understand it.
If you had RTFA, you might have noticed that they key to it working is limiting data points
If you had read my comment, you might have noticed that I said that "ideal" situations include highly dissimilar people. Having pictures that are largely full of family members is a nightmare situation for facial recognition - most algorithms have a tough enough time even figuring out where faces are, much less differentiating between kin. The scam sensor went off even more when they made a big deal about the fact that it recognizes shirt color - yeah, that's useful: If all of your pictures are of hobos.
In other words, the population it works on is always small, just the people in the pictures you have in you picture collection.
My picture collection has dozens of people that faze in and out. Again, it would be cleaning up the mess of this things inaccurate guessing more than it would be helpful.
i love to run this software over the last 6 years of digital pics i have (approx 2500 pictures) - mostly of my two kids from birth til now (damn trigger happy parents =).
it'd be very interesting to see how accurately it could pick the kids up from birth ranging through to how they look now - completely different.
if it could do it accurately, i wonder if they could use it to artificially age photographs of people - ie kids missing for years and years.
It looks like they didn't even have the name Riya as of August of this year. It was Ojos. Their main guy even has a blog where you can follow up on their Series B financing from the VCs. If this guy wants to make it big, their US and India teams should get the technology polished and then license it to Google for inclusion in their Google Desktop, with support for external media (e.g. DVDs full of photos).
Later, they could expand it out to search for the same faces in movies. Whoa! Hold on. You all are getting carried away. This is not recognizing everybody that passes by a security camera. The success of their codes depends on you having photographed the same individual multiple times. That's why facial recognition in security camera's is such a bad idea - you normally only get to train such a system on: one good photograph, or lots of terrible photographs. And then people want to pick out the bad guy as one of 5,000 people per camera per day???
If you tie this in to the running email address debate, you'll understand a little deeper. They do in fact keep track of the facial (and other) highlights, but there are too many to compare them to everyone ever scanned. Using an e-mail address significantly narrows the fields down, and then they reused the feature attributes. But otherwise, they statistically never reuse the patterns outside of your photo collection.
The Riya launch party is this Friday, November 18th in Atherton, CA. Details at http://ojos.wikispaces.com/Riya+Launch+Details
This is a nice publicity stunt, but lots of people are working on this and face recognition doesn't work very well yet. So, don't worry about your privacy just yet.
If anybody wants to invite someone who would be helpful with the testing, I'd sure love to give it a shot.
Too many photos, not enough organizational skill.
Ask and ye shall recieve, right? Please?
I remember reading an article in Wired last year about a company integrating GPS units into digital cameras. The Lat/Lon would be included into the metadata/timestamp of the image.
Think of the possibilities of this... Timestamps combined with location would make sorting thousands or even hundreds of thousands of pictures MUCH eaiser!!! Perhaps an onboard compass included would give a bearing as to which way the camera was pointing.
Does anyone else have any information about this sort of thing?
Libertas in infinitum