Immersive maps, AR search, Pixel 7, and more • The Register
Google IO Google I/O, the advertising industry’s annual developer conference, returned Wednesday to the Shoreline Amphitheater in Mountain View, Calif., for the first time in three years. The gathering remained largely a distant event due to the persistence of COVID-19, although there were enough Googlers, partners and assorted software developers in attendance to fill the site’s seats and punctuate the points. significant applause.
Sundar Pichai, CEO of Google’s parent company, Alphabet, opened the keynote by bringing up some familiar themes. He leaned into the implied sentiment “We’re here to help,” a proposition that’s increasingly uncertain in light of the many controversies facing the company.
He said he wanted to explain how Google advances its mission in two ways, “by deepening our understanding of information so that we can turn it into knowledge and by advancing the state of computing so that knowledge be easier to access, no matter who or where you are.”
The opening video delivered a more succinct version of the message: “Technology has the power to improve everyone’s life. Just build it,” was the theme.
And Google has built things, for better or for worse. Pichai announced 24 languages in Google Translate, which he attributed to advances in machine learning that can process the long tail of underrepresented languages.
“With advances in machine learning, we have developed a monolingual approach where the model learns to translate a new language without ever seeing a direct translation of it,” he said. “By collaborating with native speakers and institutions, we found that these translations were of sufficient quality to be useful.”
Pichai switched to Google Maps and described how the company uses computer vision to generate building models from satellite imagery.
“Using machine learning advances in 3D mapping, we merge billions of aerial and street-level images to create a new, high-fidelity representation of a place,” he explained. “These groundbreaking technologies come together to power a new experience in maps called immersive view. It lets you explore a place like never before.”
The video demo during the keynote showed an aerial view of the interior of a restaurant. What is remarkable about the scene is that it was not filmed using a drone, but was generated using neural network rendering software analyzing still images. The immersive view even works on mobile devices and will appear in Los Angeles, London, New York, San Francisco and Tokyo later this year, with more cities at a later date.
Google makes its Live View scene labeling technology available to ARCore developers for free through its Geospatial API. And it also extends its eco-friendly routing for Maps.
“Eco-friendly routes have already been rolled out in the United States and Canada and people have used them to travel 86 billion miles, helping to save an estimated half a million metric tons of carbon emissions, l ‘equivalent to taking 100,000 cars off the road,’ Pichai said. “I’m happy to share that we’re expanding this feature to more locations, including Europe later this year.”
AI helps video star
On YouTube, auto-generated chapters for videos are expected to grow from eight million today to 80 million over the next year. And voice recognition is applied to videos to create video transcripts which are now available for Android and iOS users.
The same goes for automatically translated captions. Pichai said machine-translated captions will be applied to Ukrainian content on YouTube next month as part of a broader effort to increase access to accurate information about the Russian-Ukrainian war.
Google’s AI recently landed in Google Docs via auto-summary. “This marks a big step forward for natural language processing,” Pichai said. “It requires understanding long passages, information compression, and language generation that were previously beyond the reach of the best machine learning models, and Docs is just the beginning.”
This tl;dr feature is now available in Spaces.
The Chocolate Factory’s fascination with AI is also evident in workspace enhancements such as “portrait light”, which will allow users of apps like Google Meet to simulate the presence of lights in the room, and “portrait restoration” to automatically improve video image quality.
To improve the presentation of various skin tones in images, Google has opened up the Monk Skin Tone Scale (MST), a framework for more accurate color rendering developed in collaboration with Harvard professor and sociologist Dr. Ellis Monk .
Prabhakar Raghavan, SVP at Google, took the stage to talk about various search improvements. The recently introduced multiple search capability – where the user takes an image and adds text to find specific information about the thing depicted – is being modified to handle the “near me” parameter, to return locally relevant results. This ability is expected to appear in English later this year.
Prabhakar Raghavan at Google IO… Click to Enlarge
Another innovation in the near future is “scene mining”, through which researchers will be able to view a scene with a mobile device camera and retrieve specific information about each element of the scene, such as the percentage of cocoa in every candy bar in the scene.
Google Assistant has learned to respond without its “Hey, Google” wake phrase. From today, the US-based Nest Hub Max can respond when looked at and addressed, for those who have opted in and pass face and voice match verifications. The device will also respond to a limited number of quick phrases, like “Set a timer for five minutes,” without “Hey, Google.”
The Android 13 beta has appeared, now in version 2. It includes a new photo picker with more granular media permissions, notification permission, and later this year will include a security settings page and unified privacy. It also comes with tablet and customization improvements.
Sameer Samat, vice president of product management at Google, highlighted Android 13’s support for RCS (Rich Communication Services), an upgrade to SMS messaging that includes end-to-end encryption. “We hope every mobile operating system will get the message and switch to RCS,” Samat said. “That way your messages are private no matter what device you’re using.”
Convenient, for safety at least, plus ear and wrist
On the hardware front, Google’s Pixel 6a will be available for pre-order, starting at $449, on July 21, with availability slated for July 28. It’s available in Chalk, Charcoal, and Sage, and shares the material used for the Pixel 6 and 6. Pro. Customers get five years of security updates, but no 3.5mm port.
Brian Rakwoski, vice president of product management, previewed the upcoming Pixel 7 and 7 Pro, due later this year.
“You can see we’ve extended the aluminum finish to the entire camera bar for the next evolution of pixel design language,” he said. “The camera body and bar are made from a single piece of 100% recycled aluminum and the beautiful pixel seven Pro and its triple camera system set a whole new standard in photography, performance and design.”
The Pixel 7, he said, will use the next-generation Google Tensor SoC and ship with Android 13.
Speaking of tensors, there was mention of eight Google TPUv4 AI Accelerator modules in an Oklahoma data center, providing approximately nine exaflops of aggregate computing capacity for Google Cloud customers so that researchers and businesses can get the same type of calculation that Google uses for its internal AI work.
There was also talk of Pixel Buds Pro, improvements to Google Wallet, and a preview of Google Pixel Watch.
Security and privacy got a lot of attention, with interface improvements like displaying account security status messages for the Google Account, expanding phishing and malware detection from Gmail to Google Docs, Sheets, and Slides, 2-Step Verification (2SV) auto-enrollment, and virtual payment cards. for Android and Chrome this summer.
Pichai concluded the keynote with a nod to augmented reality applications, such as Google Lens, multiple search, scene exploration and immersive view, as a way to enhance the real world – this which is tempting to read as a multi-billion dollar layoff of Meta CEO Mark Zuckerberg. bet on a blinded and groping virtual reality.
“That potential is what excites us most about augmented reality: the ability to spend time focusing on what matters in the real world, in our real lives,” Pichai said. “You know, the real world is pretty amazing.” ®