A world without us: can a neural network be able to remove people from photos

Not yet.

On June 24, a team of two Russian developers and an artist under the pseudonym damjanski, specializing in extravagant technology projects, released the Bye Bye Camera application. This is a camera application with a single function – removing people from the image.

The authors of Bye Bye Camera call it an application for the “post-human era”, during which cars take people away from their profession and generally put the future of the human race in question. The application is available only in the App Store and costs $2.99, but do not rush to buy it.

Although the task of removing people from an image looks trivial (anyone with a single tool in Photoshop can handle it), the neural network used by Bye Bye Camera demonstrates typical moments of frustration that can be encountered when working with any neural network. This is a low quality final result and a high error rate.

To test the Bye Bye Camera, I went to the nearest lively place – Aviapark shopping center and the park on Khodynsky field. It can be assumed that trained to remove people from an image of a neural network can help to make more clean, not loaded with photos of people in busy places – but this is not so.

Often Bye Bye Camera could not cope with single silhouettes, not to mention crowds of people and queues, but when she coped, she deleted them, leaving an extremely inaccurate background and ruining the whole composition. If people were at a distance from the lens, then the camera did not notice them, and if they were in transport — for example, on a scooter or bicycle — then the transport remained when trimming. However, it just creates the very feeling that damjanski intended.

The application description says that it “finally allows you to remove yourself from the selfie.” Well, it worked.

Back to top button