Amazon, NVIDIA and The CIA Want To Teach AI To Watch Us From Space (technologyreview.com)

An anonymous reader quotes a report from MIT Technology Review: Satellite operator DigitalGlobe is teaming up with Amazon, the venture arm of the CIA, and NVIDIA to make computers watch the Earth from above and automatically map our roads, buildings, and piles of trash. MIT Technology Review reports: "In a joint project, DigitalGlobe today released satellite imagery depicting the whole of Rio de Janeiro to a resolution of 50 centimeters. The outlines of 200,000 buildings inside the city's roughly 1,900 square kilometers have been manually marked on the photos. The SpaceNet data set, as it is called, is intended to spark efforts to train machine-learning algorithms to interpret high-resolution satellite photos by themselves. DigitalGlobe says the SpaceNet data set should eventually include high-resolution images of half a million square kilometers of Earth, and that it will add annotations beyond just buildings. DigitalGlobe's data is much more detailed than publicly available satellite data such as NASA's, which typically has a resolution of tens of meters. Amazon will make the SpaceNet data available via its cloud computing service. Nvidia will provide tools to help machine-learning researchers train and test algorithms on the data, and CosmiQ Works, a division of the CIA's venture arm In-Q-Tel focused on space, is also supporting the project." "We need to develop new algorithms for this data," says senior vice president at DigitalGlobe, Tony Frazier. He goes on to say that health and aid programs are to benefit from software that is able to map roads, bridges and various other infrastructure. The CEO of Descartes Labs, Mark Johnson, a "startup that predicts crop yields from public satellite images," says the data that is collected "should be welcome to startups and researchers," according to MIT Technology Review. "Potential applications could include estimated economic output from activity in urban areas, or guiding city governments on how to improve services such as trash collections, he says."

Comments

These are decades old computer vision projects

[perpenso]

These are decades old computer vision projects. Look through the computer vision literature going as far back as the 1980s. There are two categories that represent a large percentage of the published papers. (1) Detecting man made objects (roads, buildings, ships, vehicles, etc) from aerial and satellite imagery. (2) Detecting anomalous objects (things that don't belong there) in medical imagery.

Computers watching the earth is very old news. What is changing is that the objects being detected and described in near real-time are getting more and more complex.

Re:These are decades old computer vision projects

[Beezlebub33]

The big change here is that they are releasing marked-up data sets. That makes all the difference. A good chunk of the progress in computer vision (along with better algorithms and processing power / gpus) has been the availability of good data sets, such as ImageNet.

Machine learning algorithms, and deep learning algorithms in particular, require a lot of labeled training data. That has been largely missing from satellite imagery, for two reasons. First, nobody wanted to give up the data itself. Second, nobody wanted to go through the pain of marking up the data (by hand). This means that people that went through the bother of getting the data and labeling it (meaning large defense contractors primarily) have had a lock on wide area search, finding ships at sea, etc.

Since I don't see it, here is the link to the data on AWS: https://aws.amazon.com/public-...