I have some experience with computer vision [1]. It is non-trivial and requires real processing power to do it at scale. So I suspect they'll be charging soon unless they have a huge amount of funding. Other than that potential issue, this looks like a fairly usable, credible API.
[1] I wrote software to scan Facebook profile photos for overweight people. We store their ID's for targeting through FB ads.
I think there's an unspoken rule on the internet: any piece of personal data will eventually be used for ad targeting.
The response rates for display ads are so abysmal that any edge can make a difference. CTRs (click-through rates) hover around 0.1%, so if you increased CTR by a slim 0.1%, you'd double the revenue!
You could until a couple of months ago, and you still can with some significant caveats. Also, combining names and locations with other data from third parties gives some pretty awesome postal mail/email/telemarketing lists for companies in this space.
Yes, there are a few characteristics that are mainly visual and hard to identify any other way. Bald is one, wearing glasses (segmented by sunglasses/regular glasses) is another.
Those characteristics including ethnicity can be useful for marketing purposes.
I remember reading about a shop that used this sort of technology to profile their customers. It revealed that at a certain time of week they received an unusually large number of Japanese customers (turned out they were on a tour bus route) and responded by repositioning their Japanese member of staff to be a greeter during these periods and it significantly improved their revenue.
A similar product based around this idea might be clothes shop dummies with this built in. This would provide an interesting profile of the people interested in the particular outfit the dummy was modelling which might be useful for further advertising targeting or in the case of a multi-national brand some insights into what they should advertise in which regions.
Its open source (MIT license I believe) and performs some of the same functions as Face++ like detection, "landmarking" and replacing features/mapping a mask to your face. It includes some fun demos as well.
Looks interesting, might implement this on a project soon.
Their pricing is a bit iffy though... " The Face++ APIs are currently provided free of charge. However, Face++ may charge fees for future use of or access to the Face++ APIs or the Face++ services according to its sole discretion. If Face++ decides to charge for the Face++ APIs Services, Face++ will provide you prior notice of such charges. Face++ may also charge You when providing You a service different from the service under these Terms; for example, we may charge additional fees for excessive API use. "
not even a rough clue given as to how much they'll charge (could be reasonable, or very expensive)...
How do they get away with using all those celebrity photos?
Founder/CEO of the 3D Avatar Store here. We do most of what they do: face detection, facial feature detection and tracking, smile/expression analysis, and facial recognition. And we go further with 3D Reconstruction into performance animation ready 3D geometry, automated lip sync generation and so on.
I've tried this for a hobby project, and was quite surprised by the quality of the API. I'm a bit wary though because of the lack of a clear business model / pricing on their website.
We offer a web API that will perform 2d to 3d facial recognition which helps greatly when dealing with natural world images as opposed to mug shot type of controlled imagery. We turn the 2d image into a 3d model and then normalize the pose and lighting before performing the matching leading to higher results on everyday pictures.
Yep. I'm a 30 year old bearded man. It said it was 99.7% sure I'm a 47 year old woman. At least it got my race right! Though it was only 77% sure of that.
Recognition leaves something to be desired. Detection classifies her as female, 40 years old ±10 years. Recognition apparently doesn't add metadata parameters from detection into it's weighting:
What's the principle behind gender detection? Is it based on machine learning algorithms trained with a huge pile of data, or is there some inherent geometric difference between male and female faces that I'm missing?
There's definitely geometric differences between male and female faces. The most distinct difference is that the male chin tends to be larger, giving the face a more "square" shape, while female faces have smaller chins, giving the face a slight "almond" shape. And I'm sure there's more subtle differences as well.
Lol, it doesn't like my picture much :P I'm a 31 year old white guy (ok, was 29 in the picture) and it tells me I'm probably 43 with 7 variance and black :P
What comes to mind is a generic "upload profile photo" for websites, where the system can automatically crop your uploaded photo. Or even just to check that it is a "valid" photo.
Not sure how the gender/age info would be useful.
That's the best I can come up with. My brain actually works best in the opposite direction (problem->solution) :)
is this a project by the people at Face.com? I know they were acquired by fb some time ago so it would be neat to know they are still working on related stuff post-acquisition
You have to sign up and create an app to get the key. From what I could gather there is no limit or pricing on the site. Once you are in to the developer center they have you can see the docs and use the API. It's quite nice, even if it ages me at 31, a bit older than I am.
[1] I wrote software to scan Facebook profile photos for overweight people. We store their ID's for targeting through FB ads.