Facebook trains AI on 1 billion public Instagram photos to “see”
A person who uses Instagram.
Lorenzo Di Cola | NurPhoto via Getty Images
Pugs, Ferraris, mountains, brunches, beaches and babies – Instagram is full of them. In fact, it has become one of the largest image databases in the world over the past decade, and the company’s owner, Facebook, is using this treasure trove to teach machines what to see in a photo.
Facebook announced Thursday that it has developed an artificial intelligence program that can “see” what it sees. It did so by feeding over 1 billion public images from Instagram.
The “Computer Vision” program, nicknamed SEER, outperformed existing AI models in an object recognition test, according to Facebook.
The program achieved a “Classification Accuracy Score” of 84.2% when attempting a test of ImageNet, a large visual database developed for research into visual object recognition software. Basically, it is tested whether an AI program can recognize what can be seen in a photo.
While many AI models are trained on carefully labeled data sets, according to Facebook, SEER learned how to identify objects in photos by analyzing random, unlabeled, and unhardened Instagram images. This AI technique is known as self-supervised learning. (SEER is a piece about SElf-supERvised, according to an academic paper Facebook researchers published about the work.)
“The future of AI is to create systems that can learn directly from the information they receive – whether text, images, or any other type of data – without having to rely on carefully curated and labeled data sets to help them To teach you to recognize objects, to interpret a photo, to interpret a block of text or to do one of the myriad other tasks we ask him to do, “wrote the Facebook researchers in a blog post.
“The performance of SEER shows that self-supervised learning on computer vision tasks in real-world environments can excel,” they added. “This is a breakthrough that will ultimately pave the way for more flexible, accurate, and adaptable computer vision models in the future.”
While this is just a research project, a Facebook spokesperson said the potential uses are relatively broad. These include improved auto-generated text describing images for people with visual impairments, better auto-categorization of items sold on the Facebook marketplace, and better systems to keep harmful images off the Facebook platform, the company said.
However, many Instagram users may be surprised to hear that their images are being used to train Facebook AI systems.
“We inform Instagram account holders in our data policy that we are using the information we need to support research and innovation, including technological advances like this,” Priya Goyal, software engineer at Facebook AI Research, told CNBC.
Facebook said it will offer some of its software as an open source version so that other researchers can experiment with it.
“While we are sharing the details of our research and creating an open source library that will allow other researchers to use self-supervised learning to train models on uncured images, we are not sharing the images or the SEER mode,” Goyal said .
Other big tech companies like Google and Microsoft are also trying to push the boundaries of computer vision. Last summer, Google released the SimCLRv2 computer vision model while OpenAI released iGPT 2.