Face.com opens facial recognition API to developers | Hacker News Reader

Face.com opens facial recognition API to developers

developers.face.com

46 pointsslapshot23 comments

Who will be the first one to come out with an iPhone app that allows you to take a picture of someone and then automagically find the persons name using facial recognition?

It really seems like a pretty simple idea, especially with an API, which is already attached to Facebook, that does all of the work for you.

The API seems a bit limiting, though. For example, you must specify a list of possible users "you're looking for" in order it to find a match. The only way around this seems to be creating a private namespace that includes your own index of all Facebook users, etc.

I've implemented something like this before. I ran into a lot of performance issues though. A match against a known person seems quite fast, but when you multiply this against even a medium sized network of people the speed becomes problematic.

This was in the 90s though, I'm sure it's much faster now - I just dont know if it's fast enough to be useful.

Hey, they do it all the time on CSI and in real-time - get with the program, man!

I've mentioned making this exact thing many times to my wife & friends. Instead of iPhone, I wanted to do real-time video from a hidden cam with the processing done on a remote PC. Imagine walking around the town and casually saying "Hey Sharon" to some random girl while she and her BF both wonder how you know her name. Would be pretty funny.

I've been waiting for this to happen for a while and I'm both excited and scared.

No longer are photos anonymous just because they are in a place that doesn't have you name attached. This is just the beginning of this technology.

Suddenly blocking out faces on google steet view makes even more sense than before. I wonder if we ever get to the situation where there's a law requirement to remove faces from crowd photos in newspapers / passer-bys behind street interviews.

no reasonable expectation of privacy, not gon happen

Are you implying that there was a reasonable expectation of privacy in google steet view case?

Google did it to be nice, not because they were legally required to.

Similar API's are available all over the web. the best seems to be www.pittpatt.com but it's kinda pricey. A big advantage is that PittPatt can be installed locally on a server to provide a secure app. Privacy could be a concern as the images are stored on a CDN.

A big question is what is the confidence score of accurately matching images? The smaller the dataset the better but this was a big concern with the app that i was building. General uses like celebrity sites wont have this concern.

Interesting. How pricey?

one thing that makes me a bit nervous about this is how frequently it's wrong on its finer points. if you look at the attributes of the tagged people you'll see they are wrong about 50% of the time (for instance a smiling male is auto tagged as a non-smiling female). this makes me nervous because we are in a place now where the layman takes the absolute, digitally measured attribute as fact (ie: this is a woman, this person is smiling), instead of using their own facilities.

it's a minor thing, with this example, i know, but how long before the precision of our digital measurement ends up hurting us because the layman doesn't understand its inaccuracy?

edit: this isn't a critique of face.com, but those who may end up using it. face.com = awesome tech, very cool.

The demo page gives a lot of fun and space for experimenting... It seems popular fastfood mascots don't get recognised properly either re. sex and smiling (even though 2 of them have beards!) (tried these links: http://upload.wikimedia.org/wikipedia/en/thumb/b/bf/KFC_logo...)

yes, although i was testing their own demo pictures.

for those who downvoted: its pretty dangerous to take something that is traditionally done by humans or in an analog fashion and do it digitally when you are wrong around 50% of the time because people tend to trust technical things that look "safe enough." i understand the tech will get better, but i think its pretty bad to have attributes like "smiling?" and "male/female" even present in the API at all if i can flip a coin blindly and have roughly the same outcome.

of course its complicated, technically impressive regardless. but when a consumer sees "facial recognition technology" from the tech community they aren't about to question it. all im saying is if you are going to offer that kind of analysis, it should be right before people can start using it in their apps via an API and end consumers start depending on the data. if it's not accurate, or not a big enough a deal to care about, it shouldn't be in the API.

here come more down votes...sigh

If I remember correctly, face.com uses the technology describe in this paper: Descriptor Based Methods in the Wild. It is kind of outdated, the best algorithm nowadays can achieve around 85% accuracy in wild.

Anyone know where to find a "developer code" to sign up?

im trying to figure this out as well.

Any luck in finding this out? Google all over the place, couldn't find any reference to it.

Trying filling in everything else and leaving that field blank.

I tried a few different combos and that's that one that I think worked.

Yep. No asterisk next to the "developer code" box so it's not actually required.

Wow. The algo is really awesome. I detected every single face here: http://chir.ag/gallery/md/nature-coast-kayaking/047.jpg and got smile + glasses right for everyone except one. Nice domain too.

Does anyone know how mature algorithms that use video for facial recognition are? It seems one could use multiple consecutive frames to draw better inferences about 3D geometry, and perhaps resolve various occlusions.

uow, this is really good! Finally i can throw all my crappy opencv integrations with java :)

Reply on news.ycombinator.com