X=Likelihood of making a false match; Y=Likelihood of answering (Click to Enlarge)
Face.com is working around the clock to develop accurate and powerful face recognition technology. In our latest achievement, we’ve boosted our technology and recognition performance by approximately 30%.
We’re committed to improving face recognition technology and have published our findings in order to allow researchers from around the world to leverage our hard work for their own internal research. Therefore, we are sharing this news with the research community first, presenting preliminary metrics and visualizations at face.com/research.
We’re proud of our major improvements:
Error rates have been reduced by approximately 30%
At zero false positives (errors), we have almost doubled the best recall rate (results) to date
Many more high-confidence recognitions than before – not just saying who it is, but saying it with high confidence
Able to more accurately recognize despite aging, pose, expression and illumination factors
As part of our research, we’ve tested our new and improved algorithm with the UMass Labeled Faces in the Wild (LFW) benchmark, which is the de-facto standard test bed for unconstrained (in the wild so to speak) face recognition. We achieved a mean accuracy (average score) on the test set of 91.3% +/- 0.3. This is the best result reported to date. Many thanks for their support and making the results public.
To better explain the results, the graph (right) represents the trade-off between false-positives & true-positives (ROC curve). Interestingly, much of the improvement is achieved at the high-valued performance range – where at low false-positive rates the recall has almost doubled the best results reported to date. Trust us when we say: This is a good thing.
Our new and 30% improved algorithm will be available soon. Visit our research page at face.com/research to learn more and let us know how we can help your research by emailing us at research@face.com.
Now that over 20,000 developers are using our APIs, we wanted to create a little something extra to keep everyone in the loop. Now developers (and anyone else for that matter) can sign up to receive the Face.com Developer Newsletter. Each edition will include new features, tricks and tools that you can use in your app.
Our inaugural newsletter introduced…wait for it…Moods & Facial Expressions! Needless to day, we’re very excited as this new set of facial analysis attributes allows you to estimate moods and key facial expressions. To try these new attributes, just pass our API a photo for face detection, and there’s no need to pass any personal user identifiers. Some interesting use cases include selecting ”best” photos for display, measuring reactions on live chats, photo search filters, dating preferences and competitions.
It’s been only a few days since the release of these new features and we’ve already seen some impressive applications like this one called “Moodbattle” where you compete for who has the angriest, happiest, saddest or neutral(ist?) facial expression. Create something cool* and we’ll even send you a Face.com T-shirt. Yes, you’re welcome.
Feedback is always welcome and we look forward to seeing those API calls spike through the roof.
The Face.com Team
*Definitions of ‘cool’ may vary from individual to individual, location to location and country to country. Void where prohibited by law which is uncool.
For those of you who loved Who Framed Roger Rabbit, here’s your chance to be a Toon in Toontown. The Tides Group recently released the latest version of their Android App “Cartoonizer Free.” With over 10K downloads, it’s pretty fantastic. Cartoonizer is a photo edition tool that provides users with the capabilities to cartoonize their photos. What does it mean to ‘Cartoonzie’ a photo you say? Well, Cartoonizer uses the Face.com Face Recognition API to detect and recognize the position of the facial features of the person in the photo, and then it automatically adds the features to a cartoon image. Voila!
Face.com’s API is so powerful and accurate that it provides information on gender, head rotation in the 3 axis, whether the person is smiling, and if the individual is wearing glasses. The Cartoonizer Free app uses all this information to automate the process of cartoonizing photos. Once this cartoonization process is complete, the user can also add more features to the photo and edit all the facial features. For example, the user can edit the hue, contrast, brightness and saturation of each element, and make it match with the skin color of the person in the photo. Imagine if your bugged out Roger Rabbit-like eyes weren’t the proper shade of red and white? Not a problem with Cartoonizer. Cartoonizer is now available in the Android Market. For you, it’s free.
Thanks to Sebastian Pereyro and the The Tides Group team for demonstrating what’s possible with the Face.com API.
Recently Facebook announced their new face recognition capability was being deployed network-wide. Many news organizations picked up this story and explored the ramifications.
A recent TechCrunch article focused closely on the technology and benefits involved in using the new face recognition features. O’Reilly Radar also elaborated on how the technology is a great tool to alert people on your friend list that you might have appeared in a photo. The result means better data for the entire network in the aggregate. Another recent Forbes article also expounded on the benefits and even went so far as to explain how easy it is to change your privacy settings and disallow auto-tagging if face recognition isn’t your cup of tea.
ABC’s Nightline investigated these new features by calling on CEO Gil Hirsch to elaborate on some of the more technologically innovative and interesting aspects of face recognition. Check out the full episode here and watch the interview below.
This weekend’s TechCrunch Disrupt Hack Day in NYC saw new ideas come to life including a Face Recognition program developed by Nigel DeFreitas and Team that is a real-time personalization system. Leveraging the Face.com API, their project takes a photo of a consumer at a point of sale and then uses a recommendation engine to offer that person products and provide other useful commercial data. Once a person opts into the service, the app recognizes your face and tells you what you might want to buy based on your FB profile and then figures out what you like based on information gathered.
According to one blog, Face.com’s API shined and was in wide use during the Hackathon this weekend. Thank you to everyone who took us up on our offer of unlimited API calls. Glad we could help.
Staying up all night while eating nothing but pizza and RedBull definitely paid off. Congratulations Nigel on getting your creation covered by ZDNet and let us know how we can help in the future. That goes for all developers: Like Nigel’s Team and the others who used our Face Recognition API, we’re here to help and work with you on any new endeavors.
Face.com is here to help you code using our API. Just let us know you’re hacking for TC and we’ll provide you with unlimited calls. Check out our developer page to find out more info on using the Face.com API, and contact us with the email you used to sign up if you need to unlimit your account.
TechCrunch had this to say about this weekend’s Hackathon:
“Fueled by pizza, beer and RedBull, the hackathon event will begin Saturday afternoon through Sunday morning, May 21st – 22nd. Then after a lightning round of demos and judging Sunday afternoon, the winning team will present onstage at Disrupt on Wednesday, May 25th. This is a great opportunity for superstar hackers to share the spotlight alongside the top startups chosen from Disrupt’s Startup Battlefield.”
Android developers get your developing caps on! The challenge: Come up with a radical, mobile way to use Face.com’s face recognition API and developer tools in your next big app. Your idea could be worth $5,000 dollars in cold hard cash, a new phone and a new tablet.
Face.com and The Droid Guy are teaming up to bring you the Android Homecoming $5,000 Developer Challenge. Yes, that’s a 1/2 stack of high society. Think of all the ways you can blow 5 Grand! That’s like 100,000 bazooka joes or 1/40th of a ticket on Virgin Galactic. Either way, you’re set. But seriously, we hope you’ll use the winnings to expand upon your idea and bring it to market.
Check out entry details here. Each developer should submit a 250 word summary of their plan to incorporate the face.com face recognition API’s into their app. Please submit entries (and questions to) challenge@androidhomecoming.com no later than May 15th, 2011.
Looking forward to seeing what amazing, cool and innovative ways you leverage our API!
Good Morning America had a lot of fun seeing which famous couples look alike using Face.com’s facial recognition software. The story focuses on FindYourFaceMate.com and the theory that individuals look for mates that closely resemble themselves. Although, no one can figure out why pets often resemble their owners.
Using the Face.com API, FindYourFaceMate.com helps you find your “face mate” by matching similar features. GMA host George Stephanopoulos learned that he very closes resembles Jackie O while Cynthia McFadden was hardly upset to learn her match came up as George Clooney. Good luck Cynthia. “So you’re saying there’s a chance?”
Many thanks to GMA for featuring Face.com. Check out the whole interview here:
We love our new bigger and badder API, and we are stoked about all the ways it’s being used – from Automatic DJ’s to killer photo mashuptools. That’s why we were really excited to see two really cool and useful Face.com API applications get some attention in the past few weeks.
Shop with Your Face (literally)
First, EyeBuyDirect.com’s CEO Roy Hessel spent some time talking with the eTail blog to show off how they are using our API to personalize his customer base overnight. By connecting his database of scanned customer images with the Face.com API, he was able to identify what his customers would want to buy based on features of the image like skin tone and shape of the face – even the gender of the customer was identified with close to 95% accuracy. Read the full interview with Roy here and be sure to check out his site for awesome eyewear.
Date Your Face
Apparently, it’s a scientifically proven fact (or a really fun rumor, we’re not sure) that people are most attracted to people who resemble themselves. Armed with this organic algorithm, findyourfacemate.com harnessed our API to power their mate matching service, which is set to launch next month. Using the service, visitors will be able to upload a picture of themselves and match attributes of their photo (eyes, ears, nose, chin, etc) with a database of potential soul mates. New York Magazine went in-depth with the creator, Christina Bloom to find out how they’re using Face.com’s API and why, exactly, people would ever want to kiss someone that has their same lips. It’s a fun read about some really cool tech – check it out.
Many thanks to our die hard developer friends for coming and hanging out with Face.com on Tuesday night at Summit/IO. Face.com was thrilled to work with Summit which is an amazing incubator space for hackers, coders, developers and other techie types. We loved being a part of the scene (even for just one night) and salute you on creating an amazing, comfortable and well thought out work space.
At our shindig, new and innovative ways of using face detection were batted around, debated and explored all evening long. The hard core participants even made it out for some crazy fancy grub at Maverick in the Mission. I can still taste that sweet, sweet duck. Thanks to Summit/IO for hosting. We hope to be back again soon! Check out some more event pics below.
Follow us on Twitter to see when our next event will be and check out our new and improved Beta API. It’s quite pungent.
