When looking at photos taken from satellites, many wow how beautiful our world is. But are you sure those are real photos? Of course, the tag of a ‘satellite imagery’ makes us believe in them. But most of them are created by people who want to show our world in colors we’d like to see them in. in this regard, we even don’t want to imagine what AI can do with those images. We don’t want, but Bo Zhao from the University of Washington wants.
Deepfeakes are dangerous
Deepfakes are AI-generated synthetic imagery. But if some may use it for fun, in many other cases, it could be dangerous. Say, some people may use it for making real-looking but fake satellite imaginary. Bo Zhao and his colleagues showed the risks and opportunities involved in applying this technology to cartography. In fact, they used machine learning not for making the satellite imaginary look realistic but redrawing images in an impressionistic, crayon and arbitrary styles. However, this is clearly shows what AI can ‘make’ from satellite imaginary.
“This isn’t just Photoshopping things. It’s making data look uncannily realistic,” said Bo Zhao. “The techniques are already there. We’re just trying to expose the possibility of using the same techniques, and of the need to develop a coping strategy for it.”
How AI Can Change Satellite Imaginary?
Bo Zhao and their team trained a machine learning system on satellite images of three different cities: Seattle, Tacoma and Beijing. Each of these cities has its own distinctive look. Say, Seattle has larger overhanging greenery and narrower streets; Beijing is more monochrome. The system was trained to learn associating details of a street map from Google or Apple with those of the satellite view.
As a result, they ‘created’ a machine that returns a realistic-looking satellite image when given a street map. In other words, the agent takes the scrap of a street from the Google Map or Apple Map and shows it in the way as if it were in any those cities. Say, the map (top)corresponds to Tacoma. But the two below show how it might look in Seattle and Beijing.
Of course, the fake maps are not as sharp as the real one. Plus, when looking closely, you can see some logical inconsistencies. Don’t you see streets that go nowhere?
Where Can Deepfakes Be Used?
As the researchers prove, this technique could be used in every place where no satellite imagery is available. More interestingly, it could be used for making realistic-looking modern maps from ancient hand-drawn maps. Thus, if this technique is not used for detrimental purposes, it can be considered as a good way of creating maps of places / cities that have been ‘ignored’ by satellites.