Machine Image Completion

Creating 3D Models from Flat 2D Images
March 10, 2020
Ambient Computing Expands
March 10, 2020

Machine Image Completion

If a computer system has access to enough images it can patch and fill in holes in pictures.

If a computer system has access to enough images—millions and millions, say—it can patch and fill in holes in pictures. This capability has practical applications for professional photographers as well as everyone who wants to take a better selfie.

Soon, if the foreground of a mountain is out of focus, or if your skin has an unsightly blemish, another version can be swapped in to generate the perfect picture. But what are the next-order scenarios and implications? How will we draw the line between reality and enhancement? How much image completion should be allowed without tacking on a warning label or disclosure?

Online daters, journalists and marketers should be asking these questions. But so should policymakers.

Image completion is also a useful tool for law enforcement and military intelligence officers—computers can now assist them in identifying who or what is in the frame. Given the bias we’ve already seen across machine learning algorithms and data sets, image completion could become part of a future debate about privacy and our devices.