Facebook has announced that it is using artificial intelligence technology to ensure that 360-degree photos uploaded to social networks are of high quality.
Facebook presented a system at the scale convention that uses deep neural networks to try to solve all sorts of errors encountered in uploading photos. If the user who took the 360-degree photo can’t keep the camera perfectly in line with the horizon, there will be an image tilt, and if you watch such an image in a VR device, it will affect the immersion and the effect will be poor.
Facebook’s new system automatically corrects the horizon, so that users don’t see curved images as they look around the scene.
The new system is based on alexnet, an image recognition system that has been used to solve many other problems, such as determining the content of an image and so on. Improving the quality of 360-degree photos is valuable to Facebook, especially the virtual reality technology they invest heavily in. Facebook’s room social VR app, for example, allows avatars to be set in 360-degree photos.
But if the quality of the 360-degree photo is not good, it will greatly affect the experience of the application.
In addition to the problem of automatically correcting horizontal lines, Facebook has to deal with a large number of 360-degree photos uploaded to the server.
While this may not be a big problem in a fast network environment, it can be a problem for mobile devices in the case of cellular data. Facebook converts photos into cubes and then stores them at different resolutions. The images are then broken down into a set of 512×512 pixel squares, and Facebook can calculate the resolution of the photos and where they need to be loaded in the image. If they do not get a high enough resolution immediately, they will present a lower resolution version until they can improve the quality of the photos.