Brain-inspired artificial neural networks are at the cutting edge of AI, being used to make driverless cars speed down our roads, offer instant translations between dozens of languages, and even help with medical diagnosis.
Oh, and they can be used to help identify porn stars from a single snapshot, too.
Guessing which of these purposes the so-called Pornstar.ID (NSFW) startup falls into probably won’t come as much of a surprise. The facial recognition startup (stop snickering at the back of the class!) is one of the less likely applications of neural networks we’ve heard about, but nonetheless demonstrates that there really is no limit to where AI is popping up in our… err, some other people’s daily lives.
“Pornstar.ID is a face recognition implementation based on a deep neural network,” creator Mike Conrad told Digital Trends. “The broad workflow is as follows: first, we first detect faces with real-time pose estimation and transform the detected faces in order to make the eyes and bottom lip appear in the same location on each image. In the next step we use our deep neural network to represent the face on a 128-dimensional unit hyper sphere.”
The network, Conrad continued, was trained on upwards of 650,000 faces of more than 7,000 female adult performers. “We have an accuracy of 70 percent on 1:N and over 99 percent on 1:1 classification,” he said. At present, users can upload photos to Pornstar.ID’s website or send a tweet and then be told within seconds who that person is. If the individual isn’t known, a number of “similar” performers are listed.
“The people behind Pornstar.ID are adult webmasters from the internet’s first days, and therefore we have always been connected and interested in the adult online business,” he explained. “Together with the huge steps forward with facial recognition technology in the past few years, and the increasing demand for identification of people in adult movies, the launch of this project was a no-brainer.”
While the system currently only deals with stills, Conrad was keen to stress that the service is being expanded this year.
“We are still in beta mode and in February will launch a version with much improved version of the classifier and improved front-end,” he said. “In addition, our current private API will be opened to third parties such as tube websites, with video support, for automatic tagging of the adult performers.”
Automatic identification of individuals simply by watching videos of them? Regardless of how the technology is used, that would be a pretty exciting development!