Best Laid Plans & 3D Scans | The Juice
Zumo Labs presents The Juice, a weekly newsletter focused on computer vision problems (and sometimes just regular problems). Get it while it’s fresh.
Week of April 26–30, 2021
____
Generating synthetic datasets from CAD models or other existing 3D files is a breeze. But what if you’re starting from scratch, without access to those sorts of files? While there are online repositories of existing 3D models — Turbosquid, for example — those costs can add up, especially if you need business licensing. Instead, consider creating your own 3D model from scratch.
We just published an article on How to 3D Scan an Object for Synthetic Data. The tutorial has been provided by Sammie, 3D Artist here at Zumo Labs, and in it she walks step-by-step how to turn a physical object into the exact sort of 3D model that can form the basis of a synthetic dataset. Give it a shot. We think you’ll agree that it’s pretty amazing what you can accomplish with a cell phone camera and some open source software.
____
#Lyft
In 2016, Lyft’s president predicted that by the year 2021 a majority of the service’s rides would happen in self-driving cars. While that was overly optimistic, the company did invest a lot of time and energy building out a self-driving division of the company over the past four years. This week they announced that they would sell their entire self-driving division to Toyota’s Woven Planet. The companies expect to officially close the deal in Q3 of this year.
Lyft sells self-driving unit to Toyota’s Woven Planet for $550M, via TechCrunch.
#BadDog
The N.Y.P.D deployed their Spot robotic dog at a public housing building earlier this month, generating controversy and raising questions about the tool’s value. In the midst of the backlash, the department decided to prematurely terminate their contract with its maker, Boston Dynamics, ending the lease on April 22. Reports suggest Digidog will move to a nice farm upstate.
N.Y.P.D. Robot Dog’s Run Is Cut Short After Fierce Backlash, via The New York Times.
#TouchyFeely
If you’ve ever built furniture from Ikea, you know that sometimes you sort of need to wiggle a part to get it to fit. It turns out that building a car is no different. Robots on the Ford assembly line are now using AI technology from a firm called Symbio, which “looks at the past few hundred attempts to determine which approaches and motions appeared to work best.”
Ford’s Ever-Smarter Robots Are Speeding Up the Assembly Line, via Wired.
#FacialRecognition
India is in the midst of a full-blown Covid-19 crisis, and increasing vaccine availability is an absolute imperative. The country’s National Health Authority chief suggested that they may move to using facial recognition tech in order to authenticate individuals during vaccine distribution — a move that the Internet Freedom Foundation and other digital rights organizations have signed a statement against.
Digital Rights Bodies Warn Against Use of Facial Recognition Technology in Vaccination Drive, via The Wire.
#AJL
Kara Swisher spoke to Joy Buolamwini, founder of the Algorithmic Justice League, on a recent episode of Sway. It’s a fascinating conversation, in which they debunk the idea that any system can be “immune to algorithmic bias” and ask just how much surveillance we’re comfortable with. Can the benefits of systems like Clearview AI ever really outweigh the harms?
She’s Taking Jeff Bezos to Task, via The New York Times.
#Deepfakes
Back in 2019 the US military raised the possibility of getting bad intel in the form of geographic deepfakes. Now, a research team has published a paper showing how real that threat is. The team created software that uses GANs to generate satellite photography, and later, detection software. Incidentally, it is a process and end product that happens to show the promise of synthetic data… in this case, synthetic data’s evil twin.
Deepfake satellite imagery poses a not-so-distant threat, warn geographers, via The Verge.
____
📄 Paper of the Week
State of the Art on Neural Rendering
Neural Rendering has become a bit of a buzzword recently, and this summary paper does a really good job looking at all the different sub-fields, their history, and their future. For example, did you know that GANs are technically a type of Neural Rendering? This paper explores a wide variety of applications such as novel view synthesis, semantic photo manipulation, facial and body reenactment, relighting, free-viewpoint video, and the creation of photo-realistic avatars for virtual and augmented reality telepresence.
____
Think The Juice was worth the squeeze? Sign up here to receive it weekly.
