Google researchers know how much people like to trick others into thinking they’re on the moon, or that it’s night instead of day, and other fun shenanigans only possible if you happen to be in a movie studio in front of a green screen. So they did what any good 2018 coder would do: build a neural network that lets you do it.
This “video segmentation” tool, as they call it (well, everyone does) is rolling out to YouTube Stories on mobile in a limited fashion starting now — if you see the option, congratulations, you’re a beta tester.
A lot of ingenuity seems to have gone into this feature. It’s a piece of cake to figure out where the foreground ends and the background begins if you have a depth-sensing camera (like the iPhone X’s front-facing array) or plenty of processing time and no battery to think about (like a desktop computer).
On mobile, though, and with an ordinary RGB image, it’s not so easy to do. And if doing a still image is hard, video is even more so, since the computer has to do the calculation 30 times a second at a minimum.
Well, Google’s engineers took that as a challenge, and set up a convolutional neural network architecture, training it on thousands of labelled images like the one to the right.
The network learned to pick out the common features of a head and shoulders, and a series of optimizations lowered the amount of data it needed to crunch in order to do so. And — although it’s cheating a bit — the result of the previous calculation (so, a sort of cutout of your head) gets used as raw material for the next one, further reducing load.
The result is a fast, relatively accurate segmentation engine that runs more than fast enough to be used in video — 40 frames per second on the Pixel 2 and over 100 on the iPhone 7 (!).
This is great news for a lot of folks — removing or replacing a background is a great tool to have in your toolbox and this makes it quite easy. And hopefully it won’t kill your battery.
JD.com, one of China’s largest e-commerce companies, is launching a new Beijing-based accelerator program for artificial intelligence and blockchain startups. Called AI Catapult, its first batch includes six companies: Bankorus, CanYa, Bluezelle, Nuggets, Republic Protocol and Devery.
In an announcement, JD.com said startups will work with its operational teams to “test real-world applications of their technologies at scale.” This includes its logistics unit, which recently raised $2.5 billion and claims to run the largest last-mile logistics network in China.
Though Alibaba Group is probably better known outside of China, JD.com is a formidable rival. In 2016, it recorded 658.2 billion RMB, or about $100 billion, in gross merchandise value (JD.com will announce its full-year results for 2017 next month). JD.com and Tencent frequently partner to take on Alibaba, most recently backing several of the same online and offline retail companies, including Vipshop and Better Life. Walmart and JD.com also signed a strategic partnership in 2016 to combine their resources in China.
JD.com currently operates a sponsored AI research lab called the SAIL JD AI Research Institute with the Stanford Artificial Intelligence Laboratory. The company already uses blockchain technology in its supply chain to track products and AI in software it developed to control its logistics drones and automated package sorting centers.
Here are more details about AI Catapult’s inaugural batch:
CanYa—Based in Australia, CanYa is a peer-to-peer marketplace that lets users pay for digital or home services with cryptocurrency. During the program, CanYa will be marketed to JD.com’s customers.
Bluezelle—A Singaporean startup that provides scalable data storage and management services for decentralized apps.
Nuggets—a London-based e-commerce payments and ID platform that stores information on the blockchain to prevent data breaches.
Republic Protocol—a decentralized dark pool, or private exchange, for atomic cross-chain trading between Ether, ERC20 tokens and Bitcoin pairs.
Devery—Another Australian-based company, Devery uses blockchain tech to allow e-commerce companies to verify products through all steps of the supply chain and avoid counterfeits.
Bankorus—Formerly known as MiCai, this Chinese fintech startup claims to be the “world’s first private wealth management platform powered by AI and built on the blockchain.”
Featured Image: Bloomberg/Getty Images
In this edition of Innovate 2018, Andrew Keen finds himself in the hot seat.
Keen, whose new book, “How to Fix the Future”, was published earlier this month, discusses a moment when it has suddenly become fashionable for tech luminaries to abandon utopianism in favor of its opposite. The first generation of IPO winners have now become some of tech’s most vocal critic—conveniently of new products and services launched by a younger generation of entrepreneurs.
For example, Tesla’s Elon Musk says that advances in Artificial Intelligence present a “fundamental risk to the existence of civilization.” Salesforce CEO Marc Benioff believes Facebook ought to be regulated like a tobacco company because social media has become (literally?) carcinogenic. And Russian zillionaire George Soros last week called Google “a menace to society.”
Eschewing much of the over-the-top luddism that now fills the New York Times (“Silicon Valley is Not Your Friends”), the Guardian (“The Tech Insiders Who Fear a Smartphone Dystopia”), and other mainstream media outlets, Keen proffers practical solutions to a wide range of tech-related woes. These include persistent public and private surveillance, labor displacement, and fake news.
From experiments in Estonia, Switzerland, Singapore, India and other digital outposts, Keen distills these five tools for fixing the future:
- Increased regulation, particularly through antitrust law
- New innovations designed to solve the unintended side-effects of earlier disruptors
- Targeted philanthropy from tech’s leading moneymakers
- Modern social safety nets for displaced workers and disenfranchised consumers
- Educational systems geared for 21st century life
If you need to make a 3D model of an object, there are plenty of ways to do so, but most of them are only automated to the extent that they know how to spin in circles around that object and put together a mesh. This new system from Fraunhofer does it more intelligently, getting a basic idea of the object to be scanned and planning out what motions will let it do so efficiently and comprehensively.
It takes what can be a time-consuming step out of the process in which a scan is complete and the user has to inspect it, find where it falls short (an overhanging part occluding another, for instance, or an area of greater complexity that requires closer scrutiny) and customize a new scan to make up for these lacks. Alternatively, the scanner might already have to have a 3D model loaded in order to recognize what it’s looking at and know where to focus.
Fraunhofer’s project, led by Pedro Santos at the Institute for Computer Graphics Research, aims to get it right the first time by having the system evaluate its own imagery as it goes and plan its next move.
“The special thing about our system is that it scans components autonomously and in real time,” he said in a news release. It’s able to “measure any component, irrespective of its design — and you don’t have to teach it.”
This could help in creating one-off duplicates of parts the system has never seen before, like a custom-made lamp or container, or a replacement for a vintage car’s door or engine.
If you happen to be in Hanover in April, drop by Hannover Messe and try it out for yourself.
Featured Image: Fraunhofer
Rokid founder and CEO Mingming Zhu
Rokid, a Chinese startup that makes an AI voice assistant and smart devices, just raised a Series B extension round led by Temasek Holdings, with participation from Credit Suisse, IDG Capital and CDIB Capital. The size of the round was not released, but a source familiar with the deal told TechCrunch that it is $100 million.
The company’s previous funding was its Series B round, which was announced in November 2016. Founder and chief executive officer Mingming Zhu says Rokid raised a Series B+ instead of a C round because the company, which is based in Hangzhou, China with research centers in Beijing and San Francisco that develop its proprietary natural language processing, image processing, face recognition and robotics technology, is still in its early stages. Rokid wants to focus on gathering more resources and bringing in strategic investors like Temasek Holdings before moving on to a Series C. An investment holding company owned by the Singaporean government, Temasek Holdings counts artificial intelligence and robotics among its main investment areas and its other portfolio companies include Magic Leap.
The company’s product lineup already includes smart speakers called Rokid Pebble and Alien, which are currently sold in China. During CES, Rokid debuted its newest offering, Rokid Glass, augmented glasses created specifically for consumer use, as well as an open-source platform, called the Rokid Full Stack Open Platform. Created in partnership with Alibaba, the platform gives third-party hardware developers who use Rokid’s voice assistant access to free resources, including software blueprints and content for IoT devices. Rokid hopes that both will help build its name recognition and presence in the United States.
Reynold Wu, Rokid’s director of product management, describes the Full Stack Open Platform as a turnkey solution that not only gives developers access to Rokid’s AI technology, but also hardware solutions and services. Released with Aliyun, Alibaba’s cloud computing business, the cloud platform opened to third-party developers in China earlier this year, and will launch in the U.S. soon.
Rokid wants the platform to serve as a bridge between the two countries by giving U.S. developers an easy way to enter the Chinese market and also encouraging the development of more content for devices running Rokid’s technology, which founder and chief executive officer Mingming Zhu says is vital to attracting consumers.
“AI products are born to be global, not just for local market,” explains Zhu. “The only issue for Rokid is that we’re not ready for the U.S. market because the most important thing is content and we are not ready if there is only local content or services.”
The Pebble and Alien will be up against Google Home and Amazon Echo, which have become almost synonymous with “smart speaker” in the minds of many consumers, while Rokid Glass will inevitably be compared to Google Glass. The success of the Pebble and Alien hinge not only on how well users think Rokid’s voice assistant compares to Google Assistant and Amazon Alexa, but also the library of content and apps that the startup is able to build for its smart speakers.
While Google Glass flopped among consumers, but saw more success as an enterprise device. Rokid hopes its smart glasses, which run on its proprietary AI voice and imaging algorithms, will be able to succeed where Google Glass wasn’t because it was designed specifically for consumer applications. Early reviews from CES say the Rokid Glass is promising and praised features like face recognition, but said it still needs work to become more responsive. Once it goes on sale, the Rokid Glass will compete with smart glasses from Vuxiz, Sony and Epson. Its price hasn’t been revealed yet, but Zhu says it will be sold at a consumer-friendlier price point than its competitors (many augmented reality smart glasses from Rokid’s rivals are currently priced in the range of $600 to $1,500).
“I think we are the only product that is really consumer-centric in not only design and weight, but also energy use,” says Wu. “A lot of players design for the enterprise market first and then try consumer opportunities, but we have developed consumer products over the past three years. All of them have entered the market successfully and we have users because of that, so we have confidence in our consumer products.”
Featured Image: Rokid