AlchemyAPI, the cloud computing company that uses deep learning to analyze text and images, has added facial recognition to its portfolio of services. The company released its first computer vision service in May.
Using the company’s face API, users can now process their collections of images and tag the set of 45,000-and-growing famous people on which AlchemyAPI’s system has been trained. Users can also train the system on their own corporate networks or other internal systems in order to recognize faces within the organization. Faces the system doesn’t recognize, because it hasn’t trained on labeled images of them, still receive tags for gender and age range.
For famous people that the system recognizes, it will also provide a knowledge graph of information about the person culled from other sources on the internet.
The way the face API trains itself, AlchemyAPI Founder and CEO Elliot Turner explained, is by crawling the web and figuring out who’s being talked about a lot. After it settles on a certain list of names, another crawl goes through and searches for images by name, and disambiguates references to other people with the same name. This unsupervised approach, Turner added, is what allows AlchemyAPI to keep scaling up its operations even though it’s only a 22-person company.
It’s hard to gauge exactly how accurate the demo is across a huge spectrum of pictures, but I was impressed when Turner showed me it could distinguish actor Will Ferrell from Red Hot Chili Peppers drummer Chad Smith. He also showed identifying politicians Barack Obama, Harry Reid and Nancy Pelosi.
I ran a few more tests of the demo and found it nearly nailed my age (I’m nearly 35) and did correctly identify Microsoft Executive Vice President and General Counsel Brad Smith’s age in this photo from our Structure conference in June. He’s 55.
AlchemyAPI’s facial recognition API could not identify heavy metal star Glenn Danzig (sorry, dude) but did nail his age.
It did identify heavy metal star Dave Mustaine.
The number of faces AlchemyAPI recognizes will continue to grow, but it’s only a small step in some big new product launches the company has planned. Around November, it expects to roll out an API for recognizing the text embedded within images (for example, the sign painted on the side of a delivery van), as well as a semantic data store service called AlchemyData. Early next year, the company will launch an answers API that customers can use to ask questions of their data a la IBM’s famous Watson system.
And despite all the excitement around the deep learning models that power AlchemyAPI’s services, which has led to a spate of acquisitions of companies working on these things, Turner said his company isn’t going anywhere anytime soon. So perhaps it’s alright to get attached to the company’s services with fear they’ll be discontinued in a few months.
“Alchemy is one of the few companies in this space that’s not for sale,” he said. “… We don’t consider any acquisition offers at all.”
