Facebook’s AI revolution will start with Picasso-like photo filters

8 Nov 2016

Person taking a photo with a smartphone. Image: Yulia Grigoryeva/Shutterstock

Speaking on stage at the Web Summit in Lisbon, Facebook’s chief technology officer Mike Schroepfer laid out how the company plans to tackle AI in the coming years, starting with some smart photo filters.

While Google beat Facebook to the punch earlier this year with some incredibly smart AI capable of beating a human player at the board game Go, the social network giant has some big plans for the technology in the coming years.

So far in 2016, it has revealed how it plans to use smart chatbots to answer users’ queries in Facebook Messenger, and even use AI to act as a smart recruiter for people using its Connectivity Lab.

What is Caffe2Go?

Now Facebook’s chief technology officer has laid out how exactly the company envisions the next few years as an AI-focused business.

Speaking on stage at the Web Summit, Schroepfer explained how Facebook is attempting to build AI from the ground up, using technology like FBLearner Flow that it describes as the company’s ‘AI backbone’.

However, rather than develop technology only capable of being used on high-end desktop computers, Schroepfer has said this AI focus will start with getting the technology onto smartphones.

“This was a major engineering challenge, as we needed to design software that could run high-powered computing operations on a device with unique resource constraints in areas like power, memory, and compute capability,” he said in a blog post prior to his talk.

To that end, Facebook has now launched Caffe2Go, a new deep-learning platform developed over the past three months that can capture, analyse, and process pixels in real time on a standard iOS or Android smartphone.

Facebook AI

Facebook AI frames at work. Image: Facebook

Cramming AI into a smartphone

By condensing the size of the AI model used to process images and videos by over 100 times, Schroepfer said Caffe2Go is able to run deep neural networks with high efficiency that could be used to develop gesture-based controls for smartphone users.

Giving one example, Caffe2Go could automatically put a ‘yay’ filter over a selfie when it recognises a smile.

Given Facebook founder Mark Zuckerberg’s comments recently that it will be focusing heavily on developing video content in the coming years, it is unsurprising that AI will play an important part.

In a video of a Lisbon tram, the new filters turned it into a cubist Picasso dreamlike world, or any other famous artist’s work for that matter.

According to VentureBeat, Facebook plans to make Caffe2Go open source over the coming months to work in tandem with its existing deep-learning platform, Torch.

https://gyazo.com/836e5999c71c2330cf8376d731644b18

Colm Gorey was a senior journalist with Silicon Republic

editorial@siliconrepublic.com