Monthly Archives: January 2019

Giant artificial flowers at Google react to your emotions

Our sister team makes the machine learning-powered library driving this large installation now populating our lobby. It’s to enable this sort of thing that we released ML acceleration tech the other day:

The flowers are built using Raspberry Pi running Android Things, our Android platform for everyday devices like home speakers, smart screens and wearables. An “alpha flower” has a camera in it and uses an embedded TensorFlow neural net to analyze which emotion it sees, and the surrounding flowers change colors based on the image the camera captures of your face. All processing is done locally, so no data is saved or sent to any servers.

Better still, the code has been open-sourced.

NewImage

NewImage

Google’s using AI to help predict floods

I love the mix of high & low tech in this use of AI for social good:

A team of engineers from Google’s Crisis Response team is using machine learning to develop a more accurate and reliable flood forecasting system for countries that need it most.

Learn more about how AI and other technology is being used to help provide people with more accurate crisis alerts. http://g.co/earlywarning

NewImage

Tangentially related if at all, but what the heck: How great are these 3D renderings of vintage topographical maps?

NewImage

[YouTube]

Impressive Animoji-powered music videos

I am not, you may have noticed, curing cancer with my limited time on this planet. Having said that, I love working on the continued democratization of creative tech. These example videos show off an incredible leap in one kind of expressivity, letting one person with a telephone create animation that would’ve previously required huge amounts of effort in complex software:

Kanye/Daft Punk:

Nickelback (sorry)

Evanescence:

Accelerate machine learning through new TensorFlow Lite GPU

I’m thrilled to say that the witchcraft my team has built & used to deliver ML & AR hotness on Pixel 3, YouTube, and beyond is now available to iOS & Android developers:

For Portrait mode on Pixel 3, Tensorflow Lite GPU inference accelerates the foreground-background segmentation model by over 4x and the new depth estimation model by over 10x vs. CPU inference with floating point precision. In YouTube Stories and Playground Stickers our real-time video segmentation model is sped up by 5–10x across a variety of phones.

We found that in general the new GPU backend performs 2–7x faster than the floating point CPU implementation for a wide range of diverse deep neural network models.

A preview release is available now, with a full open source release planned for the near future.

I often note that I came here five (five!) years ago to “Teach Google Photoshop,” and delivering tech like this is a key part of that mission: enable machines to perceive the world, and eventually to see like artists & be your brilliant artistic Assistant. We have so, so far to go, and the road ahead can be far from clear—but it sure is exciting.

A little maritime drone fun

The lads and I are just back from an overnight visit to the USS Hornet, a decorated World War II-era carrier we last visited some 7 years ago. This time around we spent the night with our Cub Scout pack & several hundred other scouts & parents from around the area. On the whole we had a ball touring the ship, and I had a little fun flying my drone over the Hornet & her adjacent Navy ships:

And here’s an interactive 360º panorama from overhead. (Obligatory nerdy sidenote: This is the JPEG version stitched on the fly by the drone, and although I was able to stitch the raw source images in Camera Raw & get better color/done, I’ll be damned if I can figure out how to inject the proper metadata to make it display right. As usual I used EXIF Fixer to make the JPEG interactive.)

[YouTube]

Designers, engineers: Come work on After Effects

Here’s a rare opportunity to team up with one of the rarest of things—a super friendly, gifted, and yet humble team building a beloved app that makes the world more beautiful. The AE team have long been some of my favorite folks in the industry, and they’re looking to expand their ranks:

NewImage