Google: Our New System For Recognizing Faces Is the Best 90
schwit1 writes Last week, a trio of Google researchers published a paper on a new artificial intelligence system dubbed FaceNet that it claims represents the most accurate approach yet to recognizing human faces. FaceNet achieved nearly 100-percent accuracy on a popular facial-recognition dataset called Labeled Faces in the Wild, which includes more than 13,000 pictures of faces from across the web. Trained on a massive 260-million-image dataset, FaceNet performed with better than 86 percent accuracy.
The approach Google's researchers took goes beyond simply verifying whether two faces are the same. Its system can also put a name to a face—classic facial recognition—and even present collections of faces that look the most similar or the most distinct. Every advance in facial recognition makes me think of Paul Theroux's dystopian Ozone.
The approach Google's researchers took goes beyond simply verifying whether two faces are the same. Its system can also put a name to a face—classic facial recognition—and even present collections of faces that look the most similar or the most distinct. Every advance in facial recognition makes me think of Paul Theroux's dystopian Ozone.
Google (Score:2, Interesting)
Re:Google (Score:4, Interesting)
there was speculation that, early on, the nsa was a key funder of google. I think there was some dns registration stuff that made people do a double-take (long time ago, when google first started).
could google have gotton so far without nsa's help? one wonders. and one will never actually know, either.
Re:Google (Score:4, Interesting)
> could google have gotton so far without nsa's help? one wonders. and one will never actually know, either.
It's the other way around. NSA was interested in techniques and technology from google, especially high-performance large scale data processing. NSA was/is behind, and they knew it, and they knew the best didn't want to work with them any more when they could get a pre-IPO position at Google when Google had stunningly capable & ambitious people (2000-2005) on average.
Re: (Score:3)
and they knew the best didn't want to work with them any more
"Hey, do you want to work for us? People hate us, you'll have to pass paranoid security evaluation, but at least you'll get awesome government salary."
Re: (Score:2)
"there was speculation that, early on, the nsa was a key funder of google..."
Excellent fact based analysis. This is precisely why I come to Slashdot as opposed to some other forums frequented by conspiracy nuts.
Confusion (Score:5, Funny)
FaceNet achieved nearly 100-percent accuracy...
" performed with better than 86 percent accuracy. "
I'm not able to parse these numbers, or I have a misunderstanding as to what nearly means.
Re: (Score:3, Informative)
13,000 pictures used in Labeled Faces in the Wild test
260M pictures used in another test
Re:Confusion (Score:5, Informative)
100-percent accuracy and recognizing the presence of a face in a photo
86-percent accuracy in determining the identity of the face in the photo
Re:Confusion (Score:5, Funny)
Thanks for clearing that up.
Maybe I screwed up by being on reddit to long earlier, but unless it detects cats I see no use for this technology.
Re: (Score:2)
"Cat recognition still has a long way to go. Our best bet is to place the picture in a computer screen and use the webcam to see if the humans in front of it are amused and completely lose track of everything else"
YOu're doing it wrong, not the screen, you have to use the mouse to recognize cats.
Re: (Score:3)
100-percent accuracy and recognizing the presence of a face in a photo
86-percent accuracy in determining the identity of the face in the photo
Also two different databases. The "nearly 100%" was from a standard 13,000 photo data-set. The 86% result was from a different data-set of 260M photos. Very large training sets are critical to accurate neural net performance, so it isn't clear that they have better algorithms, rather than just more thorough training. Even using GPUs, training on that many images must have sucked up a lot of power. I am glad I don't have to pay their electric bill.
Re:Confusion (Score:4, Interesting)
The part that intrigues me is that they claim to return a name with the face.
This would imply that their facial recognition isn't just a image match, but that it looks at the context of the photos it finds to attempt to identify meta data about the people within it. Assuming that their facial recognition is no better than anyone else's recognition, by adding meta data to the calculation, especially given Google's propensity to collect and search meta data, it would seem likely that they use the meta data to make stronger identifications and find more reference photos of potential matches.
For example, if they do the first facial only search and come up with 10,000 possible matches, then they do meta searches on those 10,000 to find more pictures of them, then those pictures are compared for stronger 'training', you wind up with a much higher level of accuracy.
-Rick
Re: (Score:2)
"They probably get the name-face correlation and the 260M faces from Google+. I guess it's 26M photos from each of the ten Google+ users."
They buy access to Facebook, like everybody else and they have 7 billion photographs on picasa online albums.
Re: (Score:1)
Re: (Score:2)
"100-percent accuracy and recognizing the presence of a face in a photo"
Unfortunately they are getting _too_ good.
I noticed that with the new and improved pixel monster photos, Picasa recognized the faces of every single person in the Reunion photo behind my desk (+50 people), Beethoven's bust on the mantel and a ceramic sun dial on the wall with a face on it and the smiley on the boombox.
I guess it's time they dialed it back a little.
Re: (Score:3)
Re: (Score:1)
marketing. where a small bezel means you put "has no bezel" in big bold letters on all the marketing materials for your monitor.
Somehow I don't feel like celebrating (Score:1)
It's like Guillotine celebrating that his killing device is much more efficient than any other was in history.
It also meant that more people were decapitated than ever before.
Re: (Score:3, Funny)
Re:Somehow I don't feel like celebrating (Score:4, Insightful)
Hmm... you're right, we have to get that under control.
Get the PR dept to slap together a press kit. What we need is footage of some murderers and generally insane people with neckbeards. Make sure that you add a few shots of those islamist loonies, they're full of face wool, too. Tell them to not waste all their powder on one shot, have them release it slowly so it can sink in with the targets. Try to get that Stallman guy in there too, somehow, that way we might get a shot at that open source stuff where we can't sensibly include our backdoors. At least with the boomers it should stick, they know jack about computers but trust the TV.
Networks should gobble this up without asking, it's free news, that's all they care about.
For the millennials, try to get the story into one of those "10 things you didn't know" pages. Slap something together about 10 things you don't know about neckbeards, 10 most heinous murderers (of course you pick the neckbeard fraction), get creative! They'll love it.
Find celebrities and make sure they hate beards. Beards have to become "uncool". Only clean shaven guys get the chicks. And guys, let's not forget the fags.
Get back to me when this is rolling, maybe we don't even have to do more.
Anti-face-recognition options (Score:1)
why do we continue to do research.. (Score:4, Insightful)
on really BAD ideas and freedom-killing ideas?
you don't think this will be mostly used against us, one way or another?
do no evil? yeah, right. I have this bridge here I can sell you...
yet another example of 'lets plow ahead and not care about social blowback from our research'. scientists and engineers really need to go back to school and take an ETHICS course or two. maybe they'd realize that 'because you can, does not mean you should'.
I see nothing good coming from this. nothing at all. just pure evil to be used against people.
it takes a wise person to realize that some things should not be done. of course, google has geniuses but those geniuses have no idea at all how they are being played and how their work will be used to reduce freedom and privacy. sad that smart people can be conned into working against their own best interests ;(
Re:why do we continue to do research.. (Score:5, Funny)
why do we continue to do research..
on really BAD ideas and freedom-killing ideas?
Same reason we continue to write posts with the first sentence in the title:
Because we are fucking self-centered assholes who think we are clever and
don't give a damn whether or not our actions make life harder for other people.
Re:why do we continue to do research.. (Score:4, Insightful)
Scientists and engineers are by definition not supposed to be ethical.
You and I, as private citizens, take what they produce, determine if it's ethical/profitable/whatever, and act accordingly. Whether that is enacting a law banning said product, regulating it, or saying let the market do with it as it pleases.
As a programmer, I applaud their skill, and even more so that they were able to complete what they set out to do. As a programmer, I understand why we celebrate these type of stories.
As a private citizen, I do fear for my privacy.
But do not confuse the two perspectives.
Re: (Score:2, Insightful)
I think it is unnecessary to say the two are exclusive of each other. Of course my opinion: but I feel like scientists and engineers should have an ethical responsibility for what they create, and have the forethought and wisdom to think of its potential usages and then weigh the pros and cons.
The destruction possible through malpractice of technology, programming, and engineering is magnitudes greater than many other disciplines. That was the gist of what I learned in my ethics of computer science classes.
This is why: (Score:3)
That sounds good. The problem is, and will always be, that you may have the right ethics handled, but somewhere else, someone doesn't, and they'll just create and use the tech against you in an environment where you've not looked at it in a real world sense, and properly compensated for it to the extent that is possible.
Re: (Score:2)
An example to disprove the statement: Chemical weapons tuned for humans are a universally bad technology.
Given that the tech is possible, though, finding antidotes for them, however, is not. And making sure they work means you have to create the weapon tech as well.
So it can be about the tech; but generally speaking, it isn't.
Not sure that development of an anti-matter weapon could be justified in the context of anywhere near the planet use, either. Or black hole generation. These kinds of things are too un
Re: (Score:2)
By only considering "human", I believe you are overlooking the obvious endgame for AI. We're headed directly for the perfect race, and we're defeinitely going to get there, too. It just isn't going to be a version of us.
Re: (Score:2)
yes but Elon already warned us about that one so i think we are safe from it :)
Re: (Score:2)
You, suh, are fated to redefine "safe" :)
Re: (Score:1)
Scientists and engineers are by definition not supposed to be ethical.
That is idiotic. "Scientists and engineers" are also citizens, and have the capability and duty to think about their actions as much as anyone else.
Re:why do we continue to do research.. (Score:5, Informative)
Scientists and engineers are by definition not supposed to be ethical.
Professional Engineers (PEs) disagree:
Ethics - National Society of Professional Engineers [nspe.org]
and
National Society of Professional Engineers (NSPE) Code of Ethics for Engineers [nspe.org]
Re: (Score:2)
Yes...and? "Engineer" is a regulated term in most US states, and if you look in the codes where the rules are specified, you'll see specific exemptions for software engineers. Which is to say, they're allowed to use the term to describe themselves, but in a very real sense they are engineers in name only, since they aren't subject to any of the conditions, restrictions, or other regulations that are related to engineers.
And even if that weren't the case, there's nothing in the definition of "engineer" that
Re: (Score:3)
I disagree; by the time the thing has been developed or invented, its already too late.
inventors HAVE to be sensitive to how their ideas will be abused.
again, because it can be done, does not mean it should be done.
I don't give inventors a pass, sorry. they need to think about their actions and they should be responsible for when bad things happen directly due to their poor choices. if no one is held responsible, then bad idea will continue to enslave us.
Re: (Score:2)
Precisely.
Speak for yourself; most of us DO have ethics (Score:3)
Scientists and engineers are by definition not supposed to be ethical.
"I just invented the bomb. I didn't drop it."
--Brice, Max Headroom Episode 1 "Blipverts", 1987
Reference (in particular, the third video clip): http://www.avclub.com/article/... [avclub.com]
Back then that line was meant as tongue-in-cheek humor, funny because of its ridiculousness Depressing that we've degenerated so far that you've actually said the equivalent with all seriousness. (The same could be said for many things
Re: (Score:2)
That's not part of the definition of scientist or engineer, you drooling moron.
Re: (Score:2)
And you're too literal, you narrow minded moron.
Re: (Score:1)
So when they blur the faces on Google Street view (which many people are happy they do) it doesn't also blur out the common Colonel Sanders caricature found on the popular chicken joint.
Re: (Score:1)
Convenience trumps morality, ethics for the gross majority of the public.
Face recognition to us seems like a bad, invasive idea.. But fuck if it isn't convenient. I tiptoed in to facebook because it's the primary means of ride organization for cyclists in my area. When you're snapping photos on rides and sharing them its /great/ at recognizing people, especially if you're bad at names (I'm very very very bad at it. It takes me a good month of semi-regular social interaction to put a name to a face). It real
Re: (Score:1)
The question is "How is this a freedom killing idea." Really, how. Where's the invasion of privacy in recognizing your face. Like, you had a picture taken of you and you were just counting on people not noticing you in it, so when they made a robit that does that very well you got mad? This doesn't make any since. It's just this weird and recent iteration of paranoia that doesn't extend to actually reasonable apprehensions. My buddy is this way, it's amazingly frustrating.
Re: (Score:2)
it takes a wise person to realize that some things should not be done.
Mm...nope. Not for me. As a person who loves tech, I want to do EVERYTHING possible in tech. No matter the consequences, no matter the danger, I want to see it happen. The tech is an end in and of itself.
Re: (Score:2)
Obligatory.... (Score:3)
Human tested, (Score:1)
SkyNet approved.
but I don't WANT to get faced (Score:3)
how about instead, when Google face recognition blinks green, it provides the SSN and path to school for Sergey or Larry, instead of me. doesn't that sound appropriate?
Doppelganger (Score:2)
FR utilization (Score:2)
Services and users high quality facial recognition can be plugged into: CIA, FBI, NSA, state police, local police, bill collectors, "bounty hunters", stalkers, paparazzi, computer security, club membership and virtually every other kind of venue/door lock, real or virtual, sorting your own photos, detecting use of your face in otherwise un-discoverable venues, plus basically about any kind of surveillance you can think of.
See, here's the thing. If they have your face, and they're hunting for a match, that's
Can we get an API. (Score:3)
I wouldn't mind getting my hand on their API. I am almost done with SkyNet. I just needed better face matching... Too many accidents.
IRL Dazzle paint and face mods work (Score:1)
And so do "invisible" glasses and scatter light hoodies.
But that's internal face recoginition stats the NSA and other mil agencies don't want you to know.
Just dress up for cosplay and game events (foot whatever) and they have a really really hard time tracking and ID'ing you. Especially if you do social camoflage like changing clothing and group and walk.
I'm not sure that's a good thing (Score:1)
Real Names (Score:4, Insightful)
Now do you understand Google+'s initial policy on real names?
Pete Boyd
Scarier than humans being beaten at chess (Score:2)
Facial recognition and object recognition was always thought of in the AI community as a "pattern interpretation" skill, and we suspected that human brains have special magic gears for "effortlessly" succeeding at these sorts of tasks, while AI coders struggled to emulate our success.
Now we're seriously talking about computers already being better at these tasks than we are. This is one of those milestones in AI research when we have to cross off another item from the list of "things that keep AIs from mat
Re: (Score:2)
Don't worry, the "my brain is magic" people will come up with lots more to replace anything that's crossed off.
Now combine this with GoogleGlass... (Score:2)
Re: (Score:2)
86% (Score:2)
Just thinking about the accuracy level, there are old:middle:young, male:female, black:white:Asian etc
Can this system even tell the difference between two young white male people or two old female black people? What is the accuracy then?
Sounds pretty rubbish to me.
Google discovers Deep Learning (Score:1)
So Google has discovered Deep Learning.
I wonder if they'll publish some material claiming they invented the theory behind it in spite of all the previous work, like with their other achievements?
dopplegangers (Score:1)
Google,
can I get API access please so I can find my dopplegangers and partner(s?!)
Looking forward to a lock screen update. (Score:1)
I'll believe it when my phone can reliably recognize me.